CS Talk: Heuristic Search with Learning and its Applications to Robotics
Maxim Likhackev, Carnegie Mellon University
Abstract:
Heuristic search algorithms, such A* search, are popular means of planning for robotic systems due to the generality, simplicity and solid theoretical rigor provided by these methods. In this talk, I will first give a brief overview of the research my group have done on developing novel heuristic search algorithms that scale to real-world problems in Robotics and on using them to build planners for various robotic systems ranging from ground to aerial to mobile manipulation platforms. I will then talk in more details about one particular aspect of planning that my group has recently started to pursue, namely heuristic search-based planning with learning from experience and demonstrations. To this end, I will present several novel graph search algorithms we developed including Experience Graphs, the first to my knowledge heuristic search that improves its performance from experience, and Multi-heuristic A* search.