Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Sim and Real: Better Together
(Submitted on 1 Oct 2021 (v1), last revised 5 Oct 2021 (this version, v2))
Abstract: Simulation is used extensively in autonomous systems, particularly in robotic manipulation. By far, the most common approach is to train a controller in simulation, and then use it as an initial starting point for the real system. We demonstrate how to learn simultaneously from both simulation and interaction with the real environment. We propose an algorithm for balancing the large number of samples from the high throughput but less accurate simulation and the low-throughput, high-fidelity and costly samples from the real environment. We achieve that by maintaining a replay buffer for each environment the agent interacts with. We analyze such multi-environment interaction theoretically, and provide convergence properties, through a novel theoretical replay buffer analysis. We demonstrate the efficacy of our method on a sim-to-real environment.
Submission history
From: Shirli Di Castro Shashua [view email][v1] Fri, 1 Oct 2021 14:30:03 GMT (13429kb)
[v2] Tue, 5 Oct 2021 07:02:18 GMT (13429kb)
Link back to: arXiv, form interface, contact.