Besides the surrounding scenario understanding, the core of autonomous driving algorithms are making long-term and short-term driving policies according to the external environment and internal parameters. Autonomous driving algorithms can balance the requirement of energy consumption, time, obstacle avoidance, and comfort to find the optimized driving policy by using multi-objective optimization algorithms.
By using deep reinforcement learning, autonomous driving algorithms can learn the optimized driving policy through on-line and off-line training, and keep improving the policies by the accumulation of driving time.
The Mobility-on-Demand service provided by unmanned small buses and cars will become the main carriers in the future traffic network. A vehicle dispatching network can be scheduled by using multi-objective optimization and reinforcement learning to reduce the waiting time and to increase the vehicle utilization rate, etc.