TuSimple Develops a Long-Range Camera for Autonomous Trucks That Works at Night

author: FutureCar Staff    

While much of the auto industry is focusing on self-driving cars today, a handful of startups are working on autonomous driving systems for self-driving trucks. Among them is TuSimple, which has developed a long-range camera-based perception system tailored specifically for driverless semi trucks.

TuSimple's trucks have a vision range of 1,000 meters, which is more than any other autonomous perception system today, according to the company.

At this week's Nvidia's annual GPU Technology Conference in San Jose the company announced a low-light vision upgrade to its long-range camera perception system. The new feature will allow TuSimple's fleet of autonomous trucks to operate day or night, using a custom designed camera system.

Founded in 2015, TuSimple is developing a commercial-ready Level 4 (SAE) fully-autonomous driving solution for the logistics industry. The company is based in San Diego and is currently testing its fleet on highways in Arizona. The company plans to expand to highways in New Mexico and Texas soon.

Much of the company's focus on on its long-range perception system, which far exceeds that of autonomous cars currently being tested.

As TuSimple's entry into hardware design is a necessary one, as most cameras do not offer the performance needed for driverless trucks, which can weight up to 80,000 pounds when fully loaded. As a result, trucks take much longer to stop than a passenger car, so it is important to be able to perceive objects much further ahead in order for the autonomous driving software to react in time.

By operating at night, autonomous trucks fleets can be more efficient, increasing truck utilization from an average of 50 percent, or 12 hours per day, to an average of 80 percent.

TuSimple plans to scale its U.S. fleet to 50 trucks by June, with plans to add more. This camera system for the company's fully-autonomous Level 4 trucks is an important component of TuSimple's ability to scale, especially as the company moves beyond its validation phase.

TuSimple has leveraged Sony Semiconductor Solutions Corporation's automotive CMOS image sensor to create a set of camera systems for short, medium and long-range that use TuSimple's software to detect and process sharp images in real-time up to 1,000 meters away.

"Bringing reliable nighttime and low light operation to current perception systems in the market is an important step in the viability of autonomous driving, which is a strategic focus for us," said Tsutomu Haruta, Automotive Business Division Deputy SGM, Sony Semiconductor Solutions Corporation. "We're pleased that TuSimple, the leading self-driving truck company appreciates the cutting-edge technology that Sony is known for."

The camera system addresses complex real-world imaging challenges, such as sudden light changes when entering and exiting tunnels, glare during sunrise and sunset, and headlight glare from other vehicles. It also handles the LED flicker associated with digital signage and traffic lights.

As the system integrator, China-based Sunny Optics will develop, test and manufacture the camera module. Sunny Optics is one of the leading companies in China producing ultra-high pixel camera modules.

Hou discussed the new camera system as part of his presentation at NVIDIA's GPU Technology Conference (GTC) in San Jose, Calif., on Tuesday. The company is utilizing Nvidia's DRIVE technology for its autonomous trucks. NVIDIA is an investor in TuSimple via its venture capital arm NVIDIA GPU Ventures.

"Like human drivers, autonomous trucks' perception systems are challenged by a wide variety of light conditions which are experienced every day while driving. We weren't able to find a camera system on the market that fit our needs so we created one," said Dr. Xiaodi Hou, Founder, President and Chief Technology Officer, TuSimple. "The combined expertise of Sony Semiconductor Solutions Corporation and TuSimple has created a perception system that sees better than the human eye – night and day, rain or shine — in the most challenging driving conditions."

Along with the longe range cameras, TuSimple's self-driving trucks utilize lidar and radar to see nearly 360 degrees around for a pixel-level interpretation of the visible environment. TuSimple's autonomous hardware offers three centimeters of accuracy at all times, according to the company.

TuSimple is working to transform the $800-billion U.S. trucking industry. According to McKinsey, full autonomy for large trucks could reduce operating costs by about 45 percent, saving the trucking industry between $85 billion and $125 billion.  

With a $95 million new funding round last month, TuSimple plans to have its first commercial driverless truck operations underway by the end of 2020.

The company says its new proprietary automotive-grade camera and vision system will go into volume production in Q2 2019 and will be deployed on TuSimple's autonomous customer fleet by Q3 2019.

FutureCar Staff