37,000 Human-Planned Robotic Grasps With Six Degrees of Freedom
- 26 February 2020
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 5 (2) , 3346-3351
Abstract
Much recent work in grasp planning has focused on data-driven approaches, using deep learning to map from images to gripper configurations. However, this approach typically fails about once per ten attempts, limiting its practicality. We sought to better understand the degree to which such failures can be attributed to hardware versus control. To this end, we developed a naturalistic grasp demonstration system in which a gripper was fitted with a handle and moved by a human operator, while its trajectory was recorded with a motion tracker. The gripper's fingers were controlled with a joystick. We recorded roughly 37K grasp demonstrations with this system. These grasps were almost always successful. In contrast with planar grasp planners that perform only top-down grasps by design, many of the human-planned grasps used a horizontal approach rather than a top-down approach. We analysed robustness of a subset of these human-planned grasps, and found that many were robust to gripper rotations of about ${\pi/8}$ radians and translations of 3cm (depending on the object). Consistent with past work, human operators tended to align the gripper aperture with objects' principal axes. We also tested robustness of grasps in which the gripper aperture was aligned exactly with the principal axes, and found that these heuristic grasps were even more robust than human-planned grasps. This suggests that humans used a partly symbolic grasp planning strategy, with somewhat imprecise control.
Keywords
Funding Information
- NSERC Collaborative Research and Development
- Applied Brain Research (519891)
This publication has 18 references indexed in Scilit:
- Supersizing self-supervision: Learning to grasp from 50K tries and 700 robot hours2016
- Benchmarking in Manipulation Research: Using the Yale-CMU-Berkeley Object and Model SetIEEE Robotics & Automation Magazine, 2015
- Leveraging big data for grasp planning2015
- Deep learning for detecting robotic graspsThe International Journal of Robotics Research, 2015
- Data-Driven Grasp Synthesis—A SurveyIEEE Transactions on Robotics, 2014
- Keyframe-based Learning from DemonstrationInternational Journal of Social Robotics, 2012
- Biologically inspired grasp planning using only orthogonal approach angles2012
- Robotic Grasping of Novel Objects using VisionThe International Journal of Robotics Research, 2008
- Motor Control Strategies Revealed in the Structure of Motor VariabilityExercise and Sport Sciences Reviews, 2002
- Planning optimal grasps1992