Active Bayesian Multi-class Mapping from Range and Semantic Segmentation Observation

Abstract

Many robot applications call for autonomous exploration and mapping of unknown and unstructured environments. Information-based exploration techniques, such as Cauchy-Schwarz quadratic mutual information (CSQMI) and fast Shannon mutual information (FSMI), have successfully achieved active binary occupancy mapping with range measurements. However, as we envision robots performing complex tasks specified with semantically meaningful objects, it is necessary to capture semantic categories in the measurements, map representation, and exploration objective. This work develops a Bayesian multi-class mapping algorithm utilizing rangecategory measurements. We derive a closed-form efficiently computable lower bound for the Shannon mutual information between the multi-class map and the measurements. The bound allows rapid evaluation of many potential robot trajectories for autonomous exploration and mapping. We compare our method against frontier-based and FSMI exploration and apply it in a 3-D photo-realistic simulation environment.

Active Exploration and Mapping via Iterative Covariance Regulation over Continuous SE(3) Trajectories

“…On the other hand, the optimal path from iCR (red) tends to visit unexplored areas, by which the information quantity from EKF becomes larger. This shows the advantage of iCR-based path planning which finds the informative trajectory over the continuous control space even by starting from a random control sequence without using any high-level heuristics, such as biasing the trajectory to visit frontiers as in [10], [21], [3]. Furthermore, our simulations show that adding heuristics to the trajectory initialization in iCR can lead to more efficient exploration.…”

“…Efficient computation methods have been proposed in [3] for Causchy-Schwarz quadratic mutual information (CSQMI), and in [4] for fast Shannon mutual information (FSMI). Instead of binary occupancy grid mapping, recent information-based active mapping techniques have considered truncated signed distance field (TSDF) maps [20] and multi-category semantic maps [21]. Existing methods are, however, limited to discrete control spaces, typically arXiv:2103.05819v1 [cs.RO] 10 Mar 2021 with a finite number of possible control inputs [22], [20], [3], [4], and have not considered optimal control formulations of the active mapping problem.…”