Researchers at Stanford University recently introduced LUCIDGames, a computational technique to predict and plan adaptive trajectories for autonomous vehicles. This technique integrates an algorithm based on game theory and an estimation method.
Humans can generally figure out other drivers’ goals in their surroundings and negotiate decisions, for example, who goes first at a given intersection. In their study, Le Cleac’h and his teammates tried to replicate this ability and the complex behaviors supporting it in autonomous cars. Their ultimate goal was to allow self-driving cars to identify other vehicles’ objectives in their surroundings to plan more suitable trajectories in scenarios involving some degree of negotiation.
The technique combines two primary tools, an algorithm that relies on game theory and an estimation technique.
- The component-based on game theory allows the self-driving car to reason about the interactions with other agents (vehicles, pedestrians, etc.) when their objectives are different from its objective.
- The estimation part allows a self-driving vehicle to rapidly estimate other agents’ underlying objectives while interacting with them, for example, the desired speed, desired lane, or the level of aggressiveness of each object interacting with the self-driving vehicle.
LUCIDGames is designed to make self-driving vehicles capable of quickly identifying the objectives of both cars and pedestrians in their vicinity, even in complex scenarios.
The system consists of an “estimator,” a technique to identify drivers’ objectives, and a “decision-maker,” an algorithm that takes care of the steering angle and acceleration of the self-driving vehicle. The decision-maker determines the most suitable trajectories for the vehicle based on the information gathered by the estimator. A prediction of trajectory is made initially and then compared to what happened in reality.
After this initial training, it samples new guesses of the other agents’ trajectories close to retained guesses and evaluates their prediction performances. This process is repeated several times per refining its guess and concluding with a final prediction of how other surrounding agents will move.
With this technique, the self-driving car is also aware of when it can be confident in its guess and when the confidence is lower. It will take more cautious actions in such uncertain situations and keep greater safety distances from other vehicles.
In the future, LUCIDGames is expected to enhance the safety and reliability of self-driving vehicles. They can move in adaptive ways by anticipating the movements and actions of agents in their surroundings. So far, the team has only evaluated the technique in simulations. Now, they are also planning to test it on real autonomous cars.
The Multi-Robot Systems Lab at Stanford is experimenting on game-theoretic interactions between vehicles on small-scale model cars and a full-scale self-driving car with the Center for Automotive Research at Stanford (CARS) as partners.