The goal of this project is to make a series of manipulation algorithms for bin picking, object recognition, pose estimation, and assembly as well as a STRIPS-like task planner available as a "manipulation-in-a-box" solution that can be integrated with existing robots, such as Robonaut 2, via a ROS interface. In Phase I, this integration will be demonstrated using the Gazebo simulation environment in a manufacturing environment containing a dual-arm torso, a gantry system, and multiple 3D perception sensors, and optionally, on the official Robonaut 2 simulator, pending a software usage agreement. We will demonstrate a series of sensor-based mobile manipulation tasks such as a kitting task under varying levels of sensor noise and dynamic mission conditions. Here, the manipulation-in-a-box solution leverages the robot's on-board sensors, but performs all necessary computations with an energy footprint of less than 10W. The final deliverable is a complete manipulation solution that can be configured using a browser and interface with either a simulator or real hardware via ROS.
The manipulation-in-a-box solution is directly applicable to a multitude of NASA robotic systems such as Robonaut 2 and Valkyrie. Bin picking, object manipulation, and assembly are high-value tasks that enable the robot to perform IVA activities with and without human supervision, dramatically extending maintenance and preparation task Robonaut 2 could perform on ISS or "The Gateway".
Robotic Materials Inc. is actively marketing an autonomous, mobile manipulation solution to the manufacturing industry with an existing deployment in Denver. The proposed activities will help us to improve this product, but also pave the way fro RM Inc. to move from wheeled to humanoid robots in the future, opening up other application domains including household and elderly care.