Research paper
SHERPA · Backup-based safe exploration in simulation
SHERPA uses interval dynamics estimates and temporary backup actions to avoid perceived dangerous states. The author-institution abstract reports a simplified quadrotor simulation and a more complex aircraft-control example.
Reviewed 2026-11-05 · Source and bibliography checked
- Original title
- Safe Exploration Algorithms for Reinforcement Learning Controllers
- Publisher
- Institute of Electrical and Electronics Engineers (IEEE)
- Authors
- Tommaso Mannucci, Erik-Jan van Kampen, Cornelis de Visser, Qiping Chu
- DOI
- 10.1109/tnnls.2017.2654539
- Journal issue date
- 2018-04
- Source language
- en
- Source country
- US
- Edition
- IEEE Transactions on Neural Networks and Learning Systems
- Access
- Publisher metadata and open author manuscript
- Publication status
- Published
Source overview
SHERPA uses interval dynamics estimates and temporary backup actions to avoid perceived dangerous states. The author-institution abstract reports a simplified quadrotor simulation and a more complex aircraft-control example.
Use and review
Simulation evidence does not establish physical-robot safety. Model fidelity, constraint coverage and implementation effects require separate validation.
Sources & editions
Safe Exploration Algorithms for Reinforcement Learning Controllers ↗
Institute of Electrical and Electronics Engineers (IEEE) · IEEE Transactions on Neural Networks and Learning Systems
Author-institution metadata/abstract/DOI checked; simulations only. Perception of impendingfatalstates and intervalmodel assumptions are essential. Repository offers acceptedmanuscript; no full manuscript acquired.