Event

Lightweight Change Detection Enhances Group Decision-Making in Dynamic Environments

1 reports · 1 sources

Thiemen Siemensma and co-authors submitted a preprint paper that integrates a lightweight change detection mechanism into a Bayesian group decision-making framework: when a distribution shift is detected, robot beliefs are reset, in order to cope with dynamic environments in which the spatial correlation of tile proportions and surface patterns changes over time. The authors state that the detector requires only constant memory and constant-order per-sample computation, unlike two recent methods that need to store the full sample stream, and is therefore suitable for memory-constrained robots. Webots simulation results show that its change detection latency is competitive, its accuracy is more consistent under different communication settings, and decision accuracy improves when spatial correlation changes. All the above effects are simulation results.