Pre-registration (frozen 11 Nov 2026): is the wear cost of exploration in onboard RL schedulers specific to epsilon-greedy? Policy, task-count and noise extension of doi:10.31224/8438
Frozen protocol, decision code, simulator harness and planted-effect table for a pre-registered extension of the preprint on exploration-induced battery wear (engrXiv doi 10.31224/8438), deposited before the confirmatory run. Tests whether decaying-epsilon, UCB1 and Thompson sampling also cost wear at a fixed delivered reward, how the cost scales with the number of tasks, and its dependence on observation noise. Simulation only.
Authors
- John Goodman (ORCID: https://orcid.org/0009-0008-4080-8043)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-11
- DOI
- https://doi.org/10.5281/zenodo.24135638
- Primary Topic
- Reinforcement Learning in Robotics
- Type
- article
- Field-Weighted Citation Impact
- 0.00