Deep learning a grasp function for grasping under gripper pose uncertainty
- Submitting institution
-
Imperial College of Science, Technology and Medicine
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 2192
- Type
- E - Conference contribution
- DOI
-
10.1109/IROS.2016.7759657
- Title of conference / published proceedings
- 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems
- First page
- 4461
- Volume
- 2016-November
- Issue
- -
- ISSN
- 2153-0866
- Open access status
- Compliant
- Month of publication
- December
- Year of publication
- 2016
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
2
- Research group(s)
-
-
- Citation count
- 77
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- This Grasp Function solves two existing key problems: (i) need to sample grasp candidates (very slow grasping), (ii) inability to incorporate uncertainty in arm kinematics (inaccurate grasping). First work to achieve real-world grasping when training only in simulation, resulting in the largest grasping dataset in the world at the time. Inspired the famous Dex-Net series at UC Berkeley, currently state-of-the-art ("Notably, Johns et al. used rendered depth images with simulated noise [...] We build upon these results by [...]"). Main publication leading to John's RAEng Research Fellowship. Collaboration with Dyson that helped shape their robotics programme (Dyson, contact: FoEREF@ic.ac.uk).
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -
