Autonomous robotic manipulation with a desktop research platform

The power of robots in industry is limited both by poor usability of standard programming environments and the absence of modern perception techniques from the research community. At the same time, research progress is limited both by the difficulty of installing reliable and safe robotic systems, and by a dearth of standard benchmarks allowing a fair comparison between algorithms.

Symbio Robotics, Inc., a Bay Area startup, is building a suite of tools to address these needs. Symbio will demonstrate a benchmark research platform employing a high level, real time Python programming interface to perform unstructured bin picking. We expect this to be of interest both as a practical demonstration of Symbio’s automation capabilities, and as an invitation to researchers to use this platform to convincingly benchmark novel algorithms.

Symbio’s software provides both robot agnostic, high-level motion programming in Python, a modern language favored among researchers, as well as advanced perception capabilities. In our demo, we will show the ease and simplicity of programming by modifying robot behavior live through a Jupyter notebook. We hope this demonstration will be encouraging to anyone who has struggled to setup and work with the plethora of existing esoteric robot interfaces.

To showcase advanced perception, we will display real-time 3D point clouds obtained from a single robot-mounted RGB camera using Symbio’s structure-from-motion software. Our software computes grasps from those point clouds and perform unstructured bin picking. We expect this to excite both users who want to employ this technology and researchers who see opportunities for further development.