Russia Claims First Retrofit Driverless Truck Tech
Russian truck-maker Kamaz is claiming the world’s first retrofit autonomous driving system for heavy-duty vehicles.
The manufacturer presented teaser shots, a video clip and a brief description of the system in development dubbed Avatar. In truth, the notion of an aftermarket self-driving system has been toyed with for years. It has earlier given birth to such projects as California’s Ghost Locomotion, Comma.ai and the Canadian X Matik, to name just a few.
What’s new is that Kamaz engineers have integrated all the hardware into a single unit mounted on a cabin roof and connected to the CAN bus, including the sensors and connectivity devices such as a LiDAR, short-range and long-range cams, Wi-Fi, LTE and FM radio units. Installation is said to take only a matter of minutes. To be compatible with the driverless unit, a truck must feature an automatic transmission, electric throttle and brake controls and electric power steering connected to the CAN bus as well.
Some market-ready products follow a similar principle. One of them, LaneCruise by X Matic, a start-up founded by Tesla’s ex-design engineer Nima Ashtari, had succeeded to shorten the list of hardware to four pieces. TU-Automotive had witnessed the launch of two-piece Agropilot by Cognitive Technologies in August, 2019.
Avatar’s primary niche is harsh environments including wild fires, chemical, biological or radioactive contamination areas. For higher sturdiness, the unit is given a redundant set of key elements. Driverless mode speed is limited to 37mph, suggesting that the unit would be useless on highways.
Mount-and-go challenge
The lion share of Kamaz trucks are sold in the underdeveloped regions of the world where long lifespans of commercial vehicles and a lack of competent drivers are typical conditions. Thus, the potential demand for the aftermarket self-driving system can be good – if the company have ever coped with some specific challenges on the way to the consumers.
System developers still have to perfect the algorithms, one of the biggest challenges in the game of AV development, and to complete the physical testing. It’s safe to say that at this stage Avatar is a working prototype while a market-ready version can be years away.
The task of creating a single-unit detachable self-driving system is feasible though the user experience is unlikely to be as easy as mount-and-go, thinks Andrey Karpenkov, head of department of robotics at Kovrov State Technological Academy. A well-known fact is that machinery vision devices need calibration after mounting on the vehicle body: “The exact positions of the sensors in relation to the vehicle’s geometrical and rotational center have to be determined. This is why calibration is necessary. Moreover, in the case of insufficient firmness of the joints, calibration must be done regularly to prevent angular position error. Otherwise, the AV would wrongly define its location.”
Representatives of Zyfra Robotics, the company that has recently developed a self-driving system for the Kamaz’s serial truck models, said that their task description did not count for compatibility with a detachable platform. Such equipment would need certain maintenance to work properly, presumed its managing director Dmitriy Klebanov: “At first sight, it might require issuing formal instructions on sensor tuning, verification of working order, and calibration.”
A flight before a trip
The company statement dedicated to Avatar contained a notice that one feature under development was to accomplish a self-driving module with a quadcopter for building a digital terrain model. The research team at Innopolis University has independently tried this approach earlier. “Drone mapping for the driverless purposes has proved to be a working instrument,” said Aydar Gabdullin, research engineer in the team. “At the very least, it’s useful to take a flight over the track before actually taking the ground trip.”
“Yet, aerial mapping solutions are less reliable for this task than the ground LiDAR-based ones,” he said. “With drones, there’s a risk of incorrect processing of such objects as vertical poles, road borders and markings, detection of the nature of terrain. The error of incorrect frame matching might also occur which is not the case when you scan from the ground. For the best result, manual processing has to be done in order to prevent the risk of missing important objects.”
