VP of Engineering at TuSimple, Zehua Huang, leads the software team for TuSimple trucks. Main responsibilities within the role include core algorithms, sensor, infrastructure, map, and simulation. TuSimple is a a self-driving truck company. They dedicate their technology to provide a SAE level 4 automation for semi-trucks. TuSimple is the only self-driving truck company capable of driving from depot-to-depot without human intervention needed, which requires both complex highway and local street driving.
Zehua Huang provides “A Perspective on Sensor and Perception System for Self-Driving Trucks” at AutoSens in Detroit 2019. Sharing the sensor and perception system design philosophy for L4 semi-trucks that operate in high-speed and complex scenario as well as perform delicate manoeuvres. The unique requirements for self-driving trucks and differences between passenger vehicles will be explored in details. Zehua will present how TuSimple overcomes these challenges by developing a camera-centric perception system with the industry-leading performance. Zehua will also discuss how the overall self-driving system will be benefited from this perception design.
Zehua bestows his thoughts on sharing the roads with autonomous vehicles, who will win the race and the importance of the perception system.
What are the biggest challenges in building the core AI algorithms, infrastructure and sensor solution for autonomous driving semi-trucks?
The biggest challenges of the self-driving development is managing a process of achieving both safety and efficiency. Safety is always of paramount importance to our company, yet achieving efficiency at the same time relies on a systematic design of the development process and the usage of advanced infrastructure. We developed a lot of process and tools to enhance safety and efficiency.
What do you see as the key enabler of a safe and efficient self-driving solution?
I think key enabler is the perception system. Perception system helps self-driving vehicles understand the surrounding environment. A precise interpretation of the environment enables self-driving vehicle performing safe and efficient reaction. TuSimple developed a camera-centric perception system that interpret the surround environment as good as human. Our perception system is far-sighted, can see up to 1,000 meters with rich understandings of object’s position, speed, intention and type. This technology gives our vehicle a holistic understanding road environment. Even running in a highway speed, our vehicle has more than 30 seconds reaction time to certain road situations.
What timeframe would you put on us sharing the roads with autonomous vehicles? Do you think the timescale is different for trucks versus cars and robo-taxis?
Our self-driving trucks are operating in Arizona everyday with commercial cargo. Our technology will be ready for driverless operation on a limited number of routes in 2020-2021.
Yes, I do believe trucks will win the race in becoming fully driverless. I have two supporting reasons
- Trucks are running in a regularized operation design domain (ODD), which mainly resides in a highway. It helps reducing the complex long-tail problems and speeding up verification and validation.
- The routes are fixed prior to the mission. The one-dimensional structure is easier for map to scale up and maintain.
What are you looking forward to about presenting at AutoSens in Detroit?
I’m very excited to share our progress on perception and sensors. I will talk about the unique requirements of self-driving trucks and the differences between passenger vehicles. I will also show the performance of TuSimple camera-centric perception system.
Come to hear Zehua Huang, VP of Engineering in TuSimple providing “A Perspective on Sensor and Perception System for Self-Driving Trucks” at AutoSens. Book your tickets here >>
