Search-based Planning Laboratory researches methodologies and algorithms that enable autonomous systems to act fast, intelligently and robustly. Our research concentrates mostly on developing novel planning approaches, coming up with novel heuristic searches and investigating how planning can be combined with machine learning. Our work spans graph theory, algorithms, data structures, machine learning and of course robotics. We use our algorithms to build real-time planners for complex robotic systems operating in real world and performing challenging tasks ranging from autonomous navigation and autonomous flight to multi-agent systems and to full-body mobile manipulation.
In a bit more details, we study such problems as high-dimensional motion planning, task planning, planning under uncertainty and multi-agent planning. Our goal is to develop planners that work in real-time and deal with complex real-world environments. We are also actively pursuing planning approaches that “learn from experience”. In all of our work, we strive to develop methods that come with rigorous analytical guarantees on performance such as completeness and bounds on sub-optimality. Such guarantees help dramatically users to analyze and anticipate the behavior of autonomous systems which is crucial for safe autonomy alongside people. The lab is home to several robots including PR2 robot, segbot robot, hexarotor aerial vehicle, quadrotor aerial vehicles and few other smaller aerial vehicles. In addition, we build planners for a number of large-scale robotic systems such as humanoid robots and full-scale helicopter.
