Learning From Naturalistic Driving Data for Human-Like Autonomous Highway Driving

Driving in a human-like manner is important for an autonomous vehicle to be smart and predictable traffic participant. To achieve this goal, parameters of the motion planning module should carefully tuned, which needs great effort expert knowledge. In study, method learning cost planner from naturalistic driving data proposed. The achieved by encouraging selected trajectory approximate human under same situation. employed follows widely accepted methodology that first samples candidate trajectories space, then select one with minimal as planned trajectory. Moreover, addition traditional factors such comfort, efficiency safety, function proposed incorporate incentive behavior decision like driver, so both lane change are coupled into framework. Two types — heuristic based implemented. verify validity method, set developed using drivers collected on motorways Beijing, containing changes left right lanes, car followings. Experiments conducted respect planning, promising results achieved.