.svg)
We composed this safety report to share our organization’s commitment to safety in the design, development, testing, deployment and field monitoring of our automated driving system (ADS). While no safety system can prevent all accidents, we work hard and continuously to try to make them as infrequent as possible. Aligned with National Highway Traffic Safety Administration (NHTSA) guidance and the latest industry standards UL4600, this report demonstrates our structured approach to achieving and maintaining a high level of safety across the full life cycle of our system. We develop our safety case that is grounded in engineering rigor, organizational accountability, and a continuous improvement mindset.
The report is organized around four foundational pillars that collectively support our safety case:
1. Safety of the Architecture and Designs
Our architecture is built on the foundation of safety and security. Our architectural work begins with rigorous hazard and risk analysis, tailored specifically for operations without a human driver in the vehicle. We proactively address potential threats including but not limited to random hardware failures, functional insufficiencies, foreseeable misuses, malicious attacks, unexpected Operational Design Domain (ODD) conditions, and inherent challenges of AI performance. To ensure a robust safety net, we build in extensive redundancy across key areas including our sensor suite, computing platform, and autonomy stack, prioritizing safety over cost and schedule considerations. This multifaceted approach guarantees multiple layers of protection and fallback mechanisms to maintain safe vehicle operation and mitigate potential risks in all operating scenarios.
2. Safety of Driving Behavior
This section explains how our system behaves on the road and interacts safely with all road users. It defines the system’s ODD, and our approaches to enforce the ADS within its ODD while in operation. We address how our driving strategy ensures safe and predictable behavior in complex and dynamic environments.
3. Safety of Life-cycle Management
Safety does not end at development and deployment. This pillar covers our end-to-end process for manufacturing, testing, validation, deployment readiness and post-deployment monitoring. We describe our simulation, closed-course, and on-road testing practices, safety performance metrics, incident response protocols, and mechanisms for updating and validating the system throughout its life cycle. Our goal is to ensure that the ADS performs reliably and safely throughout all phases of its use.
4. Safety Culture
At Tensor, safety is not just a department but a core value deeply ingrained throughout the entire company, serving as our highest priority in all operations. We maintain an independent team of highly educated and experienced Systems and Safety engineers, separate from hardware and software teams, who continuously learn and rigorously adhere to industry-leading safety standards like ISO 26262, ISO 21448, and UL 4600. We foster open and direct line communication across the entire company, ensuring that safety concerns are addressed promptly and effectively. Moreover, every accident and critical near-misses are meticulously investigated with a focus on understanding root causes and implementing detailed corrective actions, while maintaining a culture of learning and improvement, free from finger-pointing. This approach underscores our commitment to continuous enhancement of our safety measures and a strong safety-first mindset at all levels.
Together, these pillars provide a transparent and comprehensive view of how safety is embedded in our automated driving system, from concept to operation. We believe this structured and holistic approach enables responsible innovation and supports public trust in automated vehicle technologies.
Safety has been integral to the design of our Generation 5 technology from its inception. We are confident that a robust autonomous driving system is built upon comprehensive design, reliable implementation, and rigorous stress testing and simulation. By integrating advanced sensors and AI, our systems achieve ultra-high resolution and long-distance perception, enabling safe decision-making in diverse scenarios and proactively avoiding hazardous situations by predicting other road users' behavior. We have implemented multiple layers of redundancy across our systems, including sensor suite, computing platform, and software stack to mitigate safety-critical faults and ensure system reliability. Safety-critical faults and failures can be mitigated by redundancy without a single point of failure. Tensor maintains stringent testing and vehicle deployment protocols to ensure every Tensor autonomous vehicle operates optimally and delivers safe, dependable transportation.
Our Generation 5 vehicle platform complies with all applicable federal, state, and local laws, as well as government standards and regulations, including Federal Motor Vehicle Safety Standards. We actively engage with local authorities, law enforcement, and local communities where Gen5 will be deployed to ensure full adherence to all operational and business practices.
A thorough hazard analysis and risk assessment (HARA) is essential for autonomous driving vehicles due to the diverse range of potential hazards. Our analysis includes consideration of random hardware failures, software development pitfalls, functional insufficiencies, foreseeable misuses, malicious attacks, challenges within the Operational Design Domain (ODD), and real-world accidents within the ODD, assuming no driver is present. We adhere to industry standards such as ISO 26262 (2018), ISO 21448 (2022), and UL4600 (2023) and monitor emerging standards like ISO/PAS 8800 (2024).
Risks associated with each hazard are categorized and prioritized, and mitigation strategies are addressed accordingly into the redundant system architecture design and fault management.
The world that our ADS navigates in is complex and dynamic. Our ADS features various sensing modalities, noted below, for both long and short range sensing, while also minimizing blind spots around the vehicle. The different sources of information provided by different sensing modalities are complementary, enabling the system to cross-verify inputs, enhance perception accuracy, and maintain robust performance under diverse environmental conditions such as low light, adverse weather, or occlusions. The perception module is one of the most important software modules of the autonomous driving system. The perception module detects, tracks and recognizes the environment around the vehicle. This redundancy and diversity in sensor data contribute significantly to the reliability and safety of the autonomous driving system.
Light Detection and Ranging (LiDAR) is an active sensor that emits laser beams. A 3D point cloud is generated by tracing the reflection of laser beams that strike objects. This is combined with the information provided by other sensors, which creates a layer of redundancy for our self-driving systems. In our Gen5 system, each vehicle is installed with two high resolution long range LiDAR on the top, and four short range wide field-of-view blindspot LiDAR on each of the four sides of the vehicle. The two long range LiDAR provide overlapping coverage in the front of the vehicle to increase the detection capability and robustness in the critical direction.
Tensor vehicles’ advanced perception module using cameras provides a robust, highly accurate object detection, tracking and classification system of the surrounding environment. The combination of high frame rates, high resolution, and high dynamic range color information makes cameras the ideal backbone for our sensor suite. Our cameras are critical for object detection, tracking, and classification, mapping, lane detection, traffic lights and signs detection, and many other tasks.
The Radar system is another sensing modality of the perception module. Radar utilizes electromagnetic waves to detect the position and speed of objects. It provides our Gen5 system an effective way of detection and tracking of dynamic objects 360 degrees around the vehicle, even in dynamic weather conditions.
In addition to recognizing the visual characteristics of emergency vehicles, detecting siren sound and the direction where it came from helps our vehicle act appropriately under emergency situations. For example, if a Tensor vehicle detects an emergency vehicle approaching from behind, it pulls over and gives way to the emergency vehicle.
Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU) are important sensors for localization in the autonomy stack. We use location signals from GNSS to initialize the localization process, and combine IMU, wheel odometry, LiDAR, camera, and other sensors together to obtain the vehicle’s position, velocity, and heading with respect to the high definition 3D prior maps.
We have designed a special sensor clearing mechanism for our sensors if they get dirty or obscured due to certain weather conditions (such as rain or fog). This mechanism can remove raindrops, fog, frost, dirt, debris, mud and even dead bugs from the sensor as quickly as possible.
Before deploying in any ODD geographical area, we collect high-definition 3D maps of the intended area to gain a full understanding of the environment, and lay a solid foundation before commencing on-road testing. Tensor’s in-house HD 3D maps provide our ADS with a deep prior knowledge of the driving environment: the topographical features of the road, the accurate 3D information of road elements, speed limits, and so on.
Our HD 3D maps are created by a scalable mapping pipeline that leverages a combination of robotics, machine learning, and humans to build and verify the maps. Our mapping pipeline has been generating whole-city scale HD 3D maps. This map is then installed on our ADS and periodically updated.
Our autonomous driving technology fully utilizes all of the information provided by the maps, while constantly sensing the environment in real time to cross-reference any changes between the real world and the map. As our fleets navigate in busy urban environments, changes are constantly ongoing. Construction, road maintenance, and sudden roadblocks are a regular scene. When such changes are detected, our ADS can plan around the scene, reroute itself, notify the rider, and alert our fleet management center to share the updated information with other vehicles in the fleet.
The Tensor XCU (Tensor Control Unit) is our central onboard computing unit of the autonomous driving technology. At every step of the entire design and development cycle of the Gen5 XCU, we followed a deliberate pursuit and uncompromising commitment to safety.
The Gen5 XCU comes with two main compute systems and one supplementary compute system. The two main systems provide high performance and redundant computing capabilities to the AI stack. The supplementary compute system has less compute, yet is at an ASIL-D safety level by the ISO 26262 standards.
In normal operation mode, the two main systems work together to provide powerful compute capabilities for the autonomy stack to utilize advanced algorithms to achieve the highest safety. The supplementary compute system serves as a monitor to watch the behavior of the two main systems. In the extreme case that both main systems fail, the supplementary compute system will take over the control of the vehicle to reach a minimal risk condition.
All our Gen5 vehicles come with redundant drive-by-wire systems. Instead of using third-party or reversed engineering solutions, we work closely with our partners to equip our ADS with reliable drive-by-wire systems designed and integrated by the vehicle manufacturer.
The redundant drive-by-wire systems provide dual braking systems, dual steering systems, and dual communication channels through two Controller Area Network (CAN) buses. If one of the braking systems fails, the other redundant braking system immediately takes over to bring the vehicle to a full stop. Two steering systems work as the redundant actuators that can fully operate in case the other one fails. Each steering system has its own drive motor system, controllers, and power supplies.
Control messages communicated to the vehicle are duplicated on separate Controller Area Network (CAN) buses. Any one path of the CAN bus failure or connector failure can be resolved by the redundant path.
Tensor has designed a fully redundant EE architecture for our ADS to mitigate hardware hazards such as power failures and overheating. It includes two independent power sources for the safety critical components. The redundant power source will ensure that our ADS can still perform critical functions upon power failures or circuit discrepancies.
COLLISION AVOIDANCE REDUNDANCY
A collision avoidance system acts as an independent safety guard, constantly scanning the road immediately ahead of, and on the side of the vehicle, avoiding a collision with other vehicles and other vulnerable road users, such as pedestrians and cyclists. This collision avoidance system serves as a safety guard in the rare event of the main autonomy system failure.
Our team of talented engineers and scientists, specialized in AI, autonomous driving, robotics, and automotive technology, have been working closely together for years to develop the autonomy stack of our self-driving vehicles. Our autonomy stack integrates numerous innovations in the fields of artificial intelligence, Vision Language Model1, robotics, electrical engineering, and vehicle engineering. The result has been a unique combination of independent research and safety-oriented engineering. With our passion and dedication leading us, we will continue to maintain our rapid progress as we take driverless vehicles from concept to reality.
Besides the above redundancy design for both software and hardware, Tensor’s autonomous driving technology is built with a fallback system to handle complicated situations without requiring human intervention. Fallback can be triggered and bring the autonomous driving vehicle to a minimum risk condition if the ADS operates beyond the operational design domain (ODD), or if ADS fails.
Each Tensor vehicle is integrated with a comprehensive vehicle health monitor that constantly scans and detects the health condition of the overall autonomous driving system. The health monitor is able to detect anomalies and failures in the following software and hardware modules of the full stack:
Tensor has followed a thorough process to identify, prioritize, and mitigate the risks of cybersecurity threats. We implement security solutions and infrastructure based on the cybersecurity principles published by NHTSA, SAE International and other international standards. Access points to our autonomous driving system are limited to what is necessary, and all access points require authentication and authorization.
We protect the cybersecurity of our ADS with layers of security protections. Safety-critical functions, such as steering, braking, and direct control of the vehicle, are strictly protected and controlled with only authorized communications. Vulnerabilities of the base vehicle are mitigated in collaboration with our OEM partners.
For example, the communications between each AV and our cloud servers are carried out via secure direct tunnels, ensuring end-to-end data integrity and confidentiality protection. Only authorized cloud services possess the cryptographic keys necessary to decrypt the transmitted packets. Only our cloud servers can decrypt and reassemble packets back into their original order.
We also carefully examine the security of our wireless communications and mitigate vulnerabilities. When operating on the road, certain systems establish communication with the vehicle via wireless cellular connections, including the fleet management system and the routing system. Our AVs are equipped with a load-balancing engine that uses bandwidth control to limit resource usage, providing reliable and stable traffic to communicate over bonded network links. To further improve our security practices, our fleet management system continuously monitors and identifies new incidents and vulnerabilities in the pattern of usage and system behavior.
As we continuously iterate and upgrade our technology, risk assessment and mitigation related to cybersecurity are built into our system design and operation. Each upgrade version undergoes extensive testing and verification processes.
At Tensor, we are focused on the full-stack Level 4 self-driving technology, achieving a driving mode-specific performance by an automated driving task2 . To be able to provide useful and convenient service to many people, we’re developing our technology to navigate urban streets in broad geographic areas of more and more cities.
A closed set of operational criteria and constraints, such as geographic area, environmental conditions, system and occupants status etc., are defined as the Operational Design Domain (ODD). The ODD is extensively tested so that we are certain that our autonomous driving system is able to operate safely within it. Different testing or deployment stages have different ODDs. In the below section, we define the ODD for our self-driving fleet.
The ODD is a set of conditions in which Tensor driverless vehicles can safely operate. The ODD consists of elements including but not limited to the geographic area of testing, road type, speed range, weather conditions, time of day, vehicle and occupants status, etc. The Tensor autonomous driving vehicle can perform dynamic driving tasks and fallbacks in the predefined ODD.
To begin with, Tensor Gen5 vehicles are designed not to operate outside our defined geographical ODD. For example, the routing system would not send any Tensor ADS to travel outside of our ODD. Before deploying in any intended ODD, we collect data in the area to gain a full understanding of the area, and lay a solid foundation before on-road testing. We thoroughly evaluate our technology performance across all of our testing modalities to ensure the intended ODD is fully verified, and that each Tensor Gen5 vehicle has undergone rigorous verification and validation testing before being deployed on the road.
In a real world scenario, when our ADS is operating within the ODD area, there could be unexpected changes that would affect the driving safety:
Sudden condition changes (such as blizzards that are not part of the defined ODD) can be detected, and our autonomous driving system is able to pull over to the side of the road ( “minimal risk condition”).
In the event of a road closure pushing the vehicle out of the geographic ODD boundary, our ADS will navigate safely with the road change and achieve the minimal risk condition near the ODD boundary if there is no other way around.
We’ve only begun on the journey of fulfilling our vision to bring safe autonomy to the world. As our technology continuously improves as we scale and gather more data, we will continue to unlock new geographical areas and operational conditions.
The Tensor ADS AI stack is meticulously designed and developed to rigorously follow the laws and rules of the road. We understand that adherence to traffic regulations is fundamental to ensuring the safety of all road users. Our system is programmed with a comprehensive understanding of traffic laws, including speed limits, lane markings, traffic signals, right-of-way rules, and other relevant regulations specific to each operational design domain (ODD).
While strict adherence to traffic laws is paramount, we recognize that certain foreseeable safetycritical situations may require nuanced behavior. For instance, navigating safely past a broken-down vehicle partially obstructing a lane might necessitate briefly nudging over double lines. In such cases, we have a well-defined process to identify, analyze, and draft behavioral exemptions. Only after thorough validation and approval of the draft behavioral exemptions, will these exemptions be enabled within the specific ODD where they are deemed necessary.
The brain inside the Tensor ADS always has a backup plan in mind. Both our perception software and our planning and control software are built with algorithm level redundancy for safety. Multiple paths of perception are connected with multiple layers of planning systems, creating a layered system fallback strategy.
Our planning module is constantly interacting with all the other AI software modules. For example, even at a protected left turn with the right-of-way to proceed, our ADS is capable of minimizing the risks potentially caused by other road users running a red light, or other unconventional behaviors.
At all times, our planning module calculates multiple potential safe trajectories (such as a safe pullover path) rendering it ever-ready to respond to an emergency. In this way, our AI software is resilient to potential software and hardware hazards by being prepared ahead of time.
Tensor built the first dedicated production facility to produce Level 4 fully driverless ADS. Purposefully designed and built by Tensor, the factory produces our advanced Gen5-powered AV fleet. The production line has completed numerous rounds of design and process optimizations to churn out Tensor’s signature ADS with an extremely high level of accuracy and consistency.
To ensure the production-level quality of the complex autonomous driving system, the production lines are equipped with a range of advanced production technologies and systems, including robots, assembly stations and conveyor systems.
We have quality checks on the assembly line for the received components and subsystems, and also for every step of the integration process when assembling the vehicles. This step-by-step verification method helps us find and eliminate defects that may arise during production and manufacturing.
Every AV coming off the end-of-line inspection then proceeds to the automatic multi-sensor calibration turntable, and goes through wheel calibration as well as temperature and waterproof testing.
%20(1).webp)
Safety is at the core of our principles and philosophy. That’s because we are committed to our mission of bringing fully autonomous vehicles to the general public and providing safer and better transportation to our cities. In order to achieve this mission, we take a hardware and software integrated approach to create a safe and reliable autonomous vehicle. Our extensive testing program covers the entire development and deployment lifecycle of the software, the hardware, and the integrated autonomous vehicle in both closed-courses and on public roads.
Our Gen5 hardware system has several thousands of components, integrating various ultrahigh resolution sensors with the Tensor Control Unit (XCU) and the vehicle.
Quality assurance on the hardware system starts from the beginning - supply chain management. Tensor works closely with our suppliers on the performance, quality, and reliability assurance on the modular level or on the component level.
Safety critical individual components go through rigorous automotive grade environmental testing. For example, the Tensor XCU goes through a Mechanical Test (shock and vibration test), Electrostatic Discharge (ESD) test, Electromagnetic Compatibility (EMC) test, and so on.
%20(1).webp)
During assembly and manufacturing, we inject testing and quality assurance at different integration steps to ensure the quality and reliability at each step of the process. We have also developed customized software and hardware tools to help maintain the consistency between every ADS.
After integration with the Gen5 system, each vehicle also undergoes automotive grade end-ofline testing, including vibration testing, waterproof testing, extensive road testing, and more.
%20(1).webp)
Software testing at Tensor is carried out at both the system level and sub-system level. The entire complex autonomy system is broken down into layers of subsystems and modules, and the interfaces and dependencies between them.
At Tensor, the first step of testing and validation takes place in a virtual world: the Tensor simulator. The Tensor simulator is a powerful multifunctional simulation and verification platform for the autonomy stack. The simulator is able to simulate any scenario tailored to the particular circumstances that the engineering team wishes to test. Our simulator can accomplish five major types of simulations:
road infrastructure and buildings
sensor and perception simulation
planning and control simulation
traffic and decision simulation
verification of the full stack autonomy software
After rigorous simulation testing, new software is tested on a closed-course test ground. During testing we evaluate whether the integrated ADS meets the performance and safety goals in constructed scenarios in the real world.
Finally, new software releases are introduced on the test fleet on the public road. We organize our ODD in multiple layers so that the released software can roll out accordingly in the different testing phases.
In the Tensor simulator, scenarios are built first by using HD 3D maps collected in the real world or new virtual cities that do not exist Within the virtual map, simulated autonomous vehicles interact with virtual environments called scenarios. The Tensor simulation platform generates three types of scenarios:
Log Simulation: simulation built from the real-world miles we drive with each autonomous driving vehicle, especially the “interesting” miles with complex or rare encounters, like a jaywalking pedestrian or a red-light runner. Our large fleet of vehicles are collecting interesting miles in various vibrant and dynamic metropolitan cities.
Virtual Simulation: our simulation experts create entirely virtual scenarios from other sources, such as public data on crash scenarios by human drivers; we also create challenging or even seemingly impossible scenarios that might have never happened in the real world, such as reckless driving.
Combination of the Log Simulation and Virtual Simulation: this is a powerful tool to expand real world scenarios into numerous sets of variations. With this tool, we could see “what would happen if it was like this” and change attributes of any object in the scenario. It enables us to simulate numerous scenarios..
The Tensor platform is fully deterministic: simulations and the results are reproducible and fully configurable. Given a set of inputs, always provides the same deterministic result. With this important and fundamental characteristic, we can test our prediction, planning and control modules independently, or together in the loop.
The simulated virtual agents in the Tensor simulator can interact with each other and the simulated Tensor vehicle. They too have “minds“ of their own and are able to follow (or not follow!) the traffic rules like a real driver, such as running red lights or not taking turns at stop signs.
Our team has programmed a realistic vehicle dynamic model that simulates the physics such as the torque output and braking force based on different throttle and brake inputs, different road conditions, and tire friction conditions. This is important for the quality of the simulation and improving the fidelity of our virtual testing compared to on-road testing.
In the Tensor simulator, 3D sensor simulations synthesize LiDAR point clouds, camera videos, and other sensor inputs. This capability enables us to bring our sensing modules, perception and localization in the loop. The entire autonomy stack can be run in the simulation platform. This is especially beneficial in testing the system performance in challenging circumstances, which may be difficult or dangerous to reproduce or repeat in reality, such as illegal driving behaviors or sudden traffic accidents. Our team can then optimize the system’s performance and test in these scenarios as many times as is necessary - all in the safety of the virtual world.
Improving the fidelity of our simulation requires minimizing the differences between the simulated and real world. Using state-of-the-art machine learning approaches, our sensor simulations are learned from the real world to create realistic synthetics.
After extensive testing and validation in simulation, we conduct real world testing in closed course facilities. Well-trained Tensor staff create a series of different scenarios mimicking real driving situations. With these tests, we test our ADS together with the base vehicle in the real world. Vehicle-in-the-loop testing in closed courses also includes automotive grade tests and durability tests, such as vibration testing and water-proof testing. Each new generation of our ADS undergoes significantly long range durability and stability tests on various road surfaces and temperature in closed courses. Only then do we conduct public road tests. A strict safety protocol is followed for both manned and unmanned testing of our ADS.
The condition of each Tensor ADS is carefully monitored and maintained. We analyze the health condition of our system at multiple layers, including inspections on the base vehicle (vehicle serviced for tire rotation, alignment, oil & air filter changes, brakes, and other car related inspections), as well as sensor suite checks, XCU and Drive-by-wire (DBW) hardware checking, software and system health checks. Some of these services are carried out by external professionals and are strictly monitored by our technicians.
Every one of our ADS goes through extensive pre-flight safety checks before getting on the road. These consist of an inspection performed by our technicians as well as pre-flight checklists performed by our experienced autonomous vehicle operators.
Tensor has a robust data logging system for data collection and analysis from real world driving experiences. With a fleet of multi-location testing and operating on the road, the data logging capability puts the rate of learning and improvement of our autonomous driving system on a flywheel.
Our data logging system is able to generate detailed datasets of an entire trip, or a certain period of time before and after certain events. In the case when our vehicle is involved in a collision, this data logging capability becomes an important tool for post-crash analysis. Following a collision, we’re able to retrieve all available data from the logs, including LiDAR point cloud and other sensor data, to evaluate factors that may have contributed to the incident. We are able to reconstruct and analyze the scenario, make appropriate updates to our system accordingly and validate the update through simulation. Thing learned by each ADS is shared with the fleet.
Over the past eight years, Tensor has tested our ADS in multiple cities globally, including the metropolitan cities around the world. Each city adds to our experience with different driving behavior, traffic rules, environments and sceneries, diverse weather, and different social interaction between road users. Testing on the public roads in each city is a critical step.
The safety of on-road testing begins with highly trained and well informed safety drivers. Tensor requires its safety drivers to complete a two-week training program tailored to ensuring proficiency in identifying problems and controlling the vehicle in a safe manner. This formal process complies with the California DMV - certified autonomous vehicle tester program. Training tools include lectures, material studies, defensive driving courses, and behind-the-wheel practice. By the end of the training, each safety driver needs to pass a qualification test which contains a written test and road test.
Tensor safety drivers understand that safety is our number one priority, and will follow our safety policy at all times while operating our autonomous vehicles. The operator is required to be attentive, and is ready to take control of the vehicle in certain circumstances, such as possible emergency situations. Their attention must be focused on the road, and they should be in position to take control of the vehicle at any moment. Test drivers are trained and required to record field data and create daily reports. These reports should reflect events completely and truthfully.
Tensor has designed an emergency protocol defining the post-crash behavior of our self-driving testing vehicles.
The Tensor autonomous system is capable of detecting that it was involved in a collision. In addition, the Tensor support team, including the roadside assistance vehicle team and the fleet management team, could also confirm whether the driverless vehicle is involved in a collision.
The Tensor support team nearby will make an initial assessment of any out of the ordinary circumstances, and call emergency services if there is an accident involving other road users or damage to public properties. The Tensor support team will be immediately dispatched to provide on-scene support, provide assistance to law enforcement, first responders, and other relevant road users in operating or interacting with the ADS. In the meantime, our fleet management team can remain in communication with anyone in or outside of the vehicle.
The Tensor ADS and supporting crew will react differently depending on the collision severity. Depending on the situation, the Tensor ADS may pull over by the side of the road, or stop at the next nearest safe parking location. In the event an airbag is deployed, the base vehicle’s engine and hybrid drive will be disabled.
From our data logging system, we are able to retrieve the data regarding the collision incident, including video and other sensor data, to reconstruct and analyze the scene and get a deeper understanding of the incident. The ADS involved in the collision will be repaired if there are damages, and thoroughly tested and validated before returning to the road.
Tensor has designed a driverless vehicle law enforcement interaction plan for the reference of law enforcement agencies. Our fully driverless vehicles are capable of detecting and recognizing both the visual and audio characteristics of police or emergency vehicles, including their siren sounds, and flashing emergency lights. After the confirmation, the ADS will safely yield the right of way to the active emergency vehicle or pullover to the side of the road safely when directed by the police.
The Tensor ADSs are designed with safety mechanisms to interact with law enforcement officers and first responders. Such mechanisms ensure that the vehicle can be immobilized (will not autonomously drive), can be disengaged from autonomous mode, can be turned off, towed, and can be accessed safely when necessary. The Tensor support team will also be notified and provide necessary assistance if such an event is detected.
Additionally, we provide detailed guidelines of communication for further engagement with law enforcement. We believe that a clear, transparent, and standardized interaction protocol for any emergency situations can prevent secondary accidents caused by miscommunication and misunderstandings.
It is of utmost importance to us that our AV be safe and courteous users of the roads. We will closely cooperate with law enforcement as we continuously test and perfect our technology.
At Tensor, safety is not merely a priority—it is a core value that guides every aspect of the organization’s operations. Leaders at all levels demonstrate a clear and sustained commitment to safety through their decisions, behaviors, and communications. This top-down leadership ensures that safety is embedded in strategic planning and daily workflows, reinforcing a culture where safety is never compromised. To uphold this commitment, we maintain an independent team of highly educated and experienced Systems and Safety engineers—separate from development teams—who continuously learn and rigorously adhere to industry-leading safety standards.
Tensor prioritizes above all the safety of everyone we share the road with. In order to achieve this, we communicate transparently with our users about our technology and how we handle emergencies. In the Tensor website for our passengers, we will provide a detailed description of the vehicle systems and how the customer should approach and interact with them. The vehicle itself provides additional instruction through clear, concise instructions.
This safety report is meant to serve as an introduction to the operation of our vehicle and provide a deeper dive into our safety systems. Tensor will continue to produce such content to share with the public.
Tensor ensures that all relevant employees take an active role to fulfill their safety duties through comprehensive training, access to resources, and ongoing support. Personnel at every level are empowered to take ownership of safety-related tasks and to intervene or raise concerns without fear of retaliation. This approach promotes a sense of responsibility, encourages proactive behavior, and ensures that safety is embedded in the day-to-day functions of every team and department.
Tensor fosters an environment in which employees feel safe and encouraged to report incidents, near misses, and potential hazards without fear of blame. Transparent processes are in place for investigating safety events and sharing findings across the organization. Lessons learned are systematically used to improve practices, procedures, and systems, turning each safety issue into an opportunity for growth and risk reduction.
Recognizing that safety is a shared responsibility, Tensor promotes collaboration across departments and disciplines. Teams across the company work together to identify risks, design safeguards, and implement solutions. This integrative approach enhances problem-solving, minimizes blind spots, and ensures that safety is considered from multiple perspectives in every project or process.
Tensor maintains a dynamic approach to safety culture, regularly evaluating its effectiveness. Areas for improvement are promptly addressed with targeted initiatives and process enhancements. By committing to ongoing assessment and evolution, Tensor ensures that its safety culture remains strong, adaptive, and aligned with best practices and regulatory standards.
The safety of our customers, of other drivers, pedestrians, and all other road users is our highest priority. This calls for a system that has safety and reliability engineered into its very fabric - a product that is engineered from the very beginning with safety in mind. In this report, we have sketched out the myriad of strategies we have undertaken to ensure the safety of our vehicles, from having high resolution sensing systems to redundancy in both software and hardware. We have made significant headway in creating robust, reliable autonomous vehicles, and we will strive for perfection in every detail. The road ahead is full of promise, and safe Tensor autonomous vehicles are the best way to reach the future.
Latest news, announcements and
updates direct to your inbox.
.webp)
.svg)
.svg)
