Volumetric Information Gain Guided Receding Horizon Planner for Active 3D Reconstruction.

Active reconstruction has been used to mobile robots, unmanned aerial vehicles (UAVs) for active perception, inspection, coverage planning, etc. Usually, it is formulated as a next-best-view (NBV) problem. However, this kind of formulation doesn't consider long-term situations, so the robots are more likely to get trapped in dead zones, or to generate local optimal paths. Besides, current methods usually are only suitable for robots with unconstrained motions and low degrees of freedom. In contrast, for robots like mobile manipulators, the problem is high dimensional and suffers from reachability constraints at the same time. Meanwhile, for reconstruction task, the continuity constraint of adjacent views should also be considered. To overcome these challenges, this paper presents a view path planner (VPP) by using forward simulation. We iteratively build a sampling exploration tree, and the sampling process is guided by the volumetric information gain computed on each branch of the tree. The reachability and continuity constraints are explicitly considered when sampling new states, which guarantees the feasibility. To solve the myopic problem for NBV-based methods, this paper presents a receding horizon planner which works like the model predictive control (MPC). Finally, we conduct comprehensive experiments over different information gain metrics and their performance are studied in detail. The results show that the proposed method achieves superior performance in terms of reconstruction efficiency and coverage rate.