We have a paper in the pipeline documenting our solution to the 2017 Mohamed Bin Zayed International Robotic Challenge. Here is the abstract:
This paper presents the system design and approach taken by the ETH team for the second challenge of the MBZIRC. We introduce our mobile manipulator mANYpulator which is developed for executing generic autonomous outdoor navigation and manipulation. The system combines a commercially available ground vehicle (Clearpath Husky) with a custom designed robot arm. Our contribution includes a complete software pipeline for autonomous map building, localization, and navigation, which relies on onboard hardware and sensors only. When deployed at the MBZIRC, the autonomous robot is required to find a tool panel on a field, pick an appropriate wrench from the panel, and operate a valve stem therewith. We demonstrate stable localization on very large outdoor areas without GPS support. Our exploration approach coupled with a path planner reliably scans the field while avoiding obstacles until the panel is found. The panel’s position is estimated from laser data within a 10 cm confidence interval. A subsequent positioning procedure achieves a precision of 4 cm. The manipulator uses series elastic actuation, making the system mechanically robust and compliant. Thanks to the precise joint torque controllability, a low end-effector impedance can be achieved and position inaccuracies during engaging with the valve can be overcome in a natural way. Furthermore, our solution incorporates a SVM based object detection method for visual servoing. This enables to locate the valve stem and wrenches on the panel with a detection rate greater than 90%. The paper discusses the performance and lessons learned while participating at the MBZIRC 2017, where we showed the successful autonomous operation during both runs of the grand challenge.
More information on this work will follow after the final version is accepted.