Autonomous Driving

AI supports autonomous-vehicle development by analyzing road data, training vehicles in simulation, and testing fixes before they reach the road.

Business impact

• Merged pull requests — GM’s VP of Autonomous Vehicles said the AI-agent workflow increased merged pull requests.
• Defects escaping later development stages — GM reported fewer defects escaping into later development stages after the workflow change.
• Completed rides — Waymo’s reported total is 20 million rides.
• Passenger capacity — Local Motors’ Olli carries up to 12 passengers.

Data requirements

• Road and vehicle data (tabular) — GM’s system analyzes road-vehicle data to find and triage problems.
• Simulation and virtual-world data (simulation) — WeRide and Waymo use virtual environments to train or evaluate autonomous driving behavior.
• Sensor data from the vehicle environment (sensor) — Waymo uses LiDAR, radar, cameras, and external audio receivers to perceive surroundings.
• Passenger dialogue and command inputs (text) — Local Motors’ Watson-based interface uses spoken and text-processing functions to interact with passengers.

AI methods and techniques

• Agent-based workflow automation — GM uses AI agents to analyze road-vehicle data, triage problems, run experiments, and test fixes.
• Simulation — Waymo and WeRide use simulation or virtual worlds to train and evaluate driving behavior before or alongside road use.
• World models — WeRide’s Genesis world model is used to train its autonomous-driving fleet in virtual environments.
• Natural language processing — Local Motors uses it for passenger interaction, recommendations, and question answering.
• Computer vision / sensor-based perception — Waymo’s vehicles combine sensors and AI to perceive the environment and navigate roads.

AI models and model families

Genesis, Waymo Driver, VectorNet, Carcraft, IBM Watson

Industries

Real-world evidence

6 documented case studies on record.

Companies using this: GM, WeRide, Local Motors, Uber, Waymo, Tesla.