Robots can be used to sample and reconstruct our environment. Measurement samples are collected from different locations so that a "distribution map" that describes certain environmental attribute (e.g., static terrain surface or spatiotemporal pollutant concentration map) can be reconstructed.
There is an urgent need to improve the operational mobility of autonomous ground vehicles to navigate in unstructured off-road environments including complex terrains such as rock fields, forests, and cross-domain areas (wet fields + vegetation regions), etc.
When uncertainty (e.g., robot's imperfect motion) is considered, stochastic methods are required to cope with the system stochasticity especially if it is time-varying due to spatiotemporal enviromental disturbances. Recent work also includes efficient reinforcement learning using minimalist training trials.
Identifying navigable space is an essential function for all robots. We develop machine learing methods that allow a robot to learn/construct navigable space from onboard perception inputs (e.g., cameras, LiDARS) and also compute feasible motion within the field of view but without the aid of a map.
When the number of robots is large, the robots need to explore 3D space and interact among themselves and also with surroundings including static obstacles and dynamic objects such as humans.
A common constraint for multi-robot system is communication, where each robot has a limited communication range and is able to exchange information only with neighbors in its vicinity. We work on designing decentralized coordination methods such as distributed task allocation mechanisms.
Copyright © 2018 - All Rights Reserved by VAIL
Template by OS Templates