DDLC Seminars - Prof. Nicholas Roy скачать в хорошем качестве

DDLC Seminars - Prof. Nicholas Roy

Title: Hybrid representations for robust and complex decision making Abstract: For many years, most robots relied on geometric models of the world for decision making, models based on point features, dense occupancy grids or action cost maps. The difficulty and cost of acquiring reliable models of the world limited the conditions under which a robot could be deployed. Modern machine learning has dramatically reduced the dependence on prior models of the world and led to remarkable changes in the capabilities of new robots, but with limitations on the robustness in the face of uncertainty, and limitations on the scalability in the face of complex planning problems. I will talk about recent work from my group in joint reasoning about hybrid representations and what these joint representations mean for planning more robust and complex missions for robots.