Human-In-The-Loop Control of a Bipedal Robot with Variable Levels of Autonomy

The goal of this project is to develop a control paradigm that allows a human/robot dyad to efficiently carry out tasks such as wayfinding and tool selection. This paradigm will combine the low-level autonomic skills of the robot, such as grasp planning and obstacle avoidance, and the high-level skills of the human, such as path planning and decision making. By capitalizing on the strengths of each partner, this paradigm should be able to minimize the workload on the human and overcome typical problems such as restricted communications bandwidth and time-delay. This paradigm will be developed and tested in parallel with Johnson Space Center's entry into the DARPA Robotics Challenge. During this period, I created the ROS package “omni_im” (https://github.com/danepowell/omni_im), which integrates the Phantom Omni with the ROS / Rviz Interactive Marker server. This allows a user to manipulate interactive markers in Rviz, such as those provided by the DRC, R2, and PR2 simulations, using the Phantom Omni. The biggest strength of the omni_im package is that it doesn’t require any special integration on the client side, so no modifications are necessary to the DRC, PR2, or R2 simulation stacks. It “just works” out-of-the-box for the most common interactive marker types. I have also created a small demo package (omni_im_demo, https://github.com/danepowell/omni_im_demo) that can be run to demonstrate the use of omni_im without installing an entire simulation stack such as the PR2. Finally, I created a video demonstrating these packages: http://www.youtube.com/watch?v=cjHVkbk8Eyg