Problem being addressed
Automated driving in urban settings is challenging mainly due to the indeterministic nature of the human participants of the traffic.
A hybrid approach that integrates a model-based path planner into a vision based framework. The agent learns to overrule the model-based planner’s decisions if it predicts that better future rewards can be obtained while doing so, e.g., avoiding an accident. During training, the driving agent is penalized for making collisions and being far from the closest waypoint asymmetrically, with the former term having precedence. This makes the agent prone to follow waypoints during free driving but gives it enough flexibility to stray from the path for collision avoidance using visual cues.
Advantages of this solution
A proof-of-concept implementation and experiments in a virtual environment showed that the proposed method is capable of learning to drive safely. The proposed method can plan its path and navigate while avoiding obstacles between randomly chosen origin-destination points.
Solution originally applied in these industries
Transport & Logistics Industry
Possible New Application of the Work
The suggested learning-based approach can be implemented in autonomous assistive surgical robots, which are gaining popularity and already outperform humans in certain types of micro surgeries.
Robotics: Potentially, the same reward strategy can be applied for integrating vehicle control and trajectory planning modules into model-free agents. Path planning and trajectory planning algorithms assume an increasing significance in Robotics; trajectory planning algorithms are crucial in Robotics, because defining the times of passage at the via-points influences not only the kinematic properties of the motion, but also the dynamic ones.
Source URL: #############