Learning from demonstration or Imitation learning is a popular approach to train the robotic agent to perform manipulation tasks.
In this research, we are recording the task demonstrations from humans in virtual reality environments as part of an episodic memory. The robot can later ask task-related questions or get training data from the task-specific motions the human has performed.
Virtual game in Unreal Engine where participants are asked to execute pouring tasks with different containers. Activity data both symbolic and sub-symbolic is logged as part of narrative-enabled episodic memory. The NEEM is then deployed on the web-based knowledge service openEASE.
PR2 performs a similar pouring action in a real apartment kitchen where the parameters for the CRAM pouring plan are derived from human VR demonstrations.
© Designed by HTML Codex
