Reference Type | Thesis |
Author(s) | Menzenbach, R. |
Year | 2019 |
Title | Benchmarking Sim-2-RealAlgorithms on Real-WorldPlatforms |
Journal/Conference/Book Title | Bachelor Thesis |
Keywords | Sim-to-real, Domain Randomization, Benchmarking |
Abstract | Learning from simulation is particularly useful, because it is typically cheaper and safer than learning on real-worldsystems. Nevertheless, the transfer of learned behavior from the simulation to the real word can impose difficultiesbecause of the so-called ’reality gap’. There are multiple approaches trying to close the gap. Although many benchmarksof reinforcement learning algorithms exist, state-of-the-art sim-2-real methods are rarely compared. In this thesis, wecompare two recent methods on Furuta pendulum swing up and ball balancing tasks. The performed benchmarks aimat assessing sim-2-sim and sim-2-real transferability. We show that the application of sim-2-real methods significantlyimproves the transferability of learned behavior. |
Date | 01.10.2019 |
Link to PDF | https://www.ias.informatik.tu-darmstadt.de/uploads/Team/FabioMuratore/Menzenbach--BenchmarkingSim2RealAlgorithmsOnRealWorldPlatforms.pdf |