An Adaptive Multi-Target RRT* Algorithm via Reachability-Aware Hierarchical Sampling

In warehousing logistics, Autonomous Mobile Robots (AMRs) are frequently tasked with visiting a dense set of target locations, such as storage bins, in a single deployment. The primary computational bottleneck in these missions lies in generating collision-free trajectories amidst narrow aisles and static racking. Conventional strategies, which either treat target connections as independent queries or initialize search forests from all targets simultaneously, often suffer from computational redundancy in such cluttered environments. In this paper, we propose the Adaptive Multi-Target RRT* (AMT-RRT*). Unlike forest-based approaches, our method maintains a single, persistent search tree rooted at the robot’s initial configuration. We introduce a Reachability-Aware Hierarchical Sampling strategy that dynamically guides the tree’s expansion through narrow passages towards the most accessible targets, thereby avoiding ineffective exploration in blocked regions. Experimental results demonstrate that this strategy significantly reduces the computation time required to connect the complete goal set compared to state-of-the-art planners.