AI-Powered Research Aims to Improve Driving Safety

Toyota Research Institute (TRI) and Stanford Engineering jointly announced a groundbreaking achievement in the field of autonomous driving research: the successful execution of tandem autonomous drifting with two vehicles.

Over the course of nearly seven years, our collaborative research efforts have been dedicated to enhancing driving safety. The experiments we have conducted involve the automation of a motorsports technique known as “drifting,” which entails precise control of a vehicle’s trajectory after losing traction by spinning the rear tires. This skill is directly applicable to real-world scenarios, such as recovering from a slide on slippery surfaces like snow or ice. By introducing a second vehicle drifting in tandem, we have effectively simulated dynamic traffic conditions that demand rapid response to the presence of other vehicles, pedestrians, and cyclists.

“Our researchers came together with one goal in mind – how to make driving safer,” said Avinash Balachandran, vice president of TRI’s Human Interactive Driving division. “Now, utilizing the latest tools in AI, we can drift two cars in tandem autonomously. It is the most complex maneuver in motorsports, and reaching this milestone with autonomy means we can control cars dynamically at the extremes. This has far-reaching implications for building advanced safety systems into future automobiles.”

“The physics of drifting are actually similar to what a car might experience on snow or ice,” said Chris Gerdes, professor of mechanical engineering and co-director of the Center for Automotive Research at Stanford (CARS). “What we have learned from this autonomous drifting project has already led to new techniques for controlling automated vehicles safely on ice.”

In a tandem drifting sequence, two vehicles, a lead car and a chase car, navigate a course in close proximity while operating at the limits of control. The team employed advanced techniques to develop the vehicle’s AI, including a neural network tire model that facilitated experiential learning, akin to that of an expert driver.

“The track conditions can change dramatically over a few minutes when the sun goes down,” said Gerdes. “The AI we developed for this project learns from every trip we have taken to the track to handle this variation.”

Road accidents are responsible for over 40,000 fatalities annually in the United States and approximately 1.35 million globally. A substantial number of these incidents occur due to drivers losing control of their vehicles in unexpected and rapidly evolving situations. Autonomous technology offers significant potential in aiding drivers to respond appropriately in such scenarios.

“When your car begins to skid or slide, you rely solely on your driving skills to avoid colliding with another vehicle, tree, or obstacle. An average driver struggles to manage these extreme circumstances, and a split second can mean the difference between life and death,” added Balachandran. “This new technology can kick in precisely in time to safeguard a driver and manage a loss of control, just as an expert drifter would.”

“Doing what has never been done before truly shows what is possible,” added Gerdes. “If we can do this, just imagine what we can do to make cars safer.”

Technical Details

  • Experiments were conducted at Thunderhill Raceway Park in Willows, California, utilizing two modified GR Supras. Algorithms for the lead car were developed by TRI, while Stanford engineers developed those for the chase car.
  • TRI concentrated on developing robust and stable control mechanisms for the lead car, enabling it to perform repeatable, safe lead runs.
  • Stanford Engineering developed AI vehicle models and algorithms that allow the chase car to adapt dynamically to the lead car’s motion, enabling it to drift alongside without collision.
  • Toyota Racing Development (TRD) and GReddy modified the suspension, engine, transmission, and safety systems (e.g., roll cage, fire suppression) of each vehicle. Although subtly different, the vehicles were built to the same specifications used in Formula Drift competitions to facilitate data collection with expert drivers in a controlled environment.
Both vehicles are equipped with computers and sensors that enable them to control their steering, throttle, and brakes while also sensing their motion (e.g., position, velocity, and rotation rate).

  • Importantly, the vehicles share a dedicated WiFi network, enabling real-time communication and information exchange, such as their relative positions and planned trajectories.
  • To achieve autonomous tandem drifting, the vehicles must continuously plan their steering, throttle, and brake commands, as well as the trajectory they intend to follow, using a technique known as Nonlinear Model Predictive Control (NMPC).

In Model Predictive Control (NMPC), each vehicle begins with objectives, mathematically represented as regulations or limitations, that it must adhere to.

  • The primary vehicle's objective is to maintain a controlled drift along a designated path while adhering to the constraints imposed by the laws of physics and hardware limitations, such as the maximum steering angle.
  • The secondary vehicle's objective is to drift alongside the primary vehicle while actively avoiding a collision.

Each vehicle then solves and re-solves an optimization problem at a high frequency to determine the optimal steering, throttle, and brake commands that align with its objectives while adapting to dynamic conditions.

By utilizing AI to continuously train the neural network with data gathered from previous tests, the vehicles demonstrate progressive improvement with each trip to the track.