A Toyota Research Institute (TRI) and Stanford University have been working together for almost seven years now in a project aimed at making driving safer. Their special focus is perfecting the emergency response of self-driving cars, which they do with an extremely spectacular method: they taught the autopilot to drift!
For this, it is necessary to continuously steer the vehicle that is on the verge of spinning back into a stable, controlled driving position, which means that the vehicle control can correct the situation threatening to slide in a fraction of a second. At the beginning of 2022, the whole world marveled at the self-driving Toyota GR Supra, which was able to maintain a slide in autonomous mode, just like a professional pilot.
The collaboration has now entered a new chapter: two cars were taken to coordinated drifting at the same time. This is an exponentially more complex task than simple drifting, as the car must pay attention to its surroundings in addition to itself – in this respect, it is much closer to the challenges that await self-driving cars in traffic.
The complexity of the task is indicated by the fact that although the hardware of the two cars was almost completely identical, their controls and sensors were programmed for completely different tasks. The car in front was prepared for the maneuver by TRI researchers: its task was to maintain its stability while following the designated track and not exceed certain parameters (e.g. steering angle). The following Supra was programmed by Stanford specialists: for the car in front he had to adjust his movement, always making sure that even in the event of an unexpected situation, he could avoid a collision with a safe margin.
Following the principle of Non-Linear Model Predictive Control (NMPC), the cars continuously re-evaluated the optimal steering angles, throttle position and brake pressure up to 50 times per second. The two cars were connected to a common Wi-Fi network, through which they continuously shared their own data with each other, from the current position to the designated route.
As before, learning technology was used in this phase of development as well, so the cars will become more skillful and experienced with each experiment.
Source: www.vezess.hu
