Kodiak boosts self driving redundancy

  • April 17, 2023
  • William Payne

Kodiak Robotics has boosted sensor redundancy and GPU processing in the latest release of its autonomous truck hardware platform. The latest version of the platform, its fifth, removes the roof-mounted “centre pod” sensor suite. It also relocates the front-facing Luminar Iris LiDAR and wide field-of-view camera to each of Kodiak’s proprietary mirror-mounted SensorPods.

The company claims it has made customer upfitting faster, and the platform more maintainable and serviceable. The new release also doubles the platform’s LiDAR coverage at long range.

By relocating the sensors to the SensorPods located on the side-mounted mirrors, Kodiak has simplified the build process. It claims it has also improved the maintainability of the system. By having sensors located in an easy-to-reach location on the mirrors, customers avoid having to access the roof to maintain sensors.

Kodiak’s SensorPods are also designed to place sensors at the same height as a driver, maximizing road safety and improving perception. The SensorPods offer a better dual vantage point, as they provide redundancy and visibility on either side of the truck, as opposed to the single vantage point above the cab. In the event of a sensor damage, such as a rock or other piece of debris damaging a sensor, offering dual-redundant front-facing sensors increases resiliency. An on-the-fly replacement of SensorPods, which are compatible with any truck platform, can be performed in as little as ten minutes, maximizing uptime and customers’ revenue generation due to increased asset utilization. The damaged SensorPod can then be returned to Kodiak for evaluation and repair.

As it has removed the centre pod, the company has added a second forward-facing LiDAR and additional camera to add redundancy in the left and right SensorPods. The new truck increases the total number of sensors on-board from 14 to 18, including one new LiDAR and three new cameras—bringing the total camera count to 10. Two wide-angle cameras have been added to the hood-mounted mirrors to cover blind-spots. With the new sensor additions, Kodiak fifth-generation truck remains the most lean and fully-redundant platform in the industry.

The new truck includes the recently announced Ambarella CV2 perception system-on-chip (SoC) which handles all camera data processing. The Ambarella CV2 SoC improves image quality for longer range detections and unlocks improved dynamic range for nighttime driving.

“Customers and OEMs aren’t sensor experts, and they shouldn’t have to be, which is why they constantly tell us they love our modular approach that solves one of the biggest barriers to servicing and maintaining autonomous trucks,” said Don Burnette, Founder and CEO of Kodiak Robotics. “We took a platform that our customers already love and made it better by adding more visibility, more power, and more flexibility, ultimately moving us closer down the path towards driverless deployment. By removing the sensors from the top of the truck and incorporating them at a human driver’s line of sight, we have designed a system for the real world. Additionally, having sensors on top of the cab is actually very difficult to service while on the road and requires specialized equipment, which is nearly impossible to find roadside.”

Kodiak has also reduced the electrical power requirements for its fifth-generation truck, while simultaneously improving the processing power of the system. In addition to 130% more GPU processing power, the new system provides 60% more central processing power, and additional system redundancy. The reduced power consumption allowed for a 50% reduction in the size of the fifth-generation power system, and also decreased cooling needs.

Kodiak has taken a modular approach to building its computing system. It has simplified the integration of the system into the truck and made it easier to manage when trucks are incorporated into a customer’s fleet. Kodiak has consolidated its in-house compute system, networking, and power distribution into one physical unit. This can be built to accommodate the specifications of each customer, giving more options for integration. The compute hardware is manufactured by Crystal Group, which manufactures military and ruggedized computers. The hardware is designed to handle the heavy and persistent vibrations onboard long-haul trucks.

The company recently announced a 24-month United States Department of Defense (DoD) project to help automate future US Army ground vehicles. Advances made within the scope of the project have served as the foundation for the fifth-generation truck.