9 May 2012 Fast online learning of control regime transitions for adaptive robotic mobility
Author Affiliations +
We introduce a new framework, Model Transition Control (MTC), that models robot control problems as sets of linear control regimes linked by nonlinear transitions, and a new learning algorithm, Dynamic Threshold Learning (DTL), that learns the boundaries of these control regimes in real-time. We demonstrate that DTL can learn to prevent understeer and oversteer while controlling a simulated high-speed vehicle. We also show that DTL can enable an iRobot PackBot to avoid rollover in rough terrain and to actively shift its center-of-gravity to maintain balance when climbing obstacles. In all cases, DTL is able to learn control regime boundaries in a few minutes, often with single-digit numbers of learning trials.
© (2012) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Brian Yamauchi, Brian Yamauchi, } "Fast online learning of control regime transitions for adaptive robotic mobility", Proc. SPIE 8387, Unmanned Systems Technology XIV, 838709 (9 May 2012); doi: 10.1117/12.919444; https://doi.org/10.1117/12.919444


Back to Top