Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulation
An AI model was trained on data from an expert driver to predict vehicle responses and inputs, creating a control model that allows a vehicle to autonomously drift around obstacles.
“what they did is they had a car and they drove it around. It was super instrumented, um, and they had this expert drifter, Ken Gucci, drive it around. And I didn't know who Ken Gushi was, but I looked him up and he's got, like, probably one of the most impressive drifting driver resumes.”
Engineering & simulationnonlinear model predictive control (NMPC) with AI model
The system used an AI model to generate expert driver inputs and ran forecasts on vehicle states 20 times per second to apply computer-controlled steering, throttle, and braking on a Toyota Supra.
“So step one, it understands the situation. Step 2 it uses the AI model to generate what would an expert driver do if they were in this in this situation. Step 3 is to run forecasts on the effects to the future state based on the AI model. So remember, we said the AI model is also well versed in the outputs, and then step 4 is to actually go apply that to the system.”
Engineering & simulationTechnical University of Munich autonomous racing system
The team developed an AI system that processes sensor data (lasers, cameras, radar) to predict the trajectory of other cars and compute optimal paths for autonomous racing at high speeds.
“So it has whatever map they're racing in. It has a digital understanding of that layout. But what he's really trying to do is predict the trajectory of the cars around it that it's racing with as precisely as possible so that he can find the most optimum path for the next movement to make, right?”