Sampling-based Motion Planning for Robotic Information Gathering

Abstract

Abstract-We propose an incremental sampling-based motion planning algorithm that generates maximally informative trajectories for guiding mobile robots to observe their environment. The goal is to find a trajectory that maximizes an information metric (e.g., variance reduction, information gain, or mutual information) and also falls within a pre-specified budget constraint (e.g., fuel, energy, or time). Prior algorithms have employed combinatorial optimization techniques to solve these problems, but existing techniques are typically restricted to discrete domains and often scale poorly in the size of the problem. Our proposed Rapidly-exploring Information Gathering (RIG) algorithm extends ideas from Rapidly-exploring Random Graphs (RRGs) and combines them with branch and bound techniques to achieve efficient optimization of information gathering while also allowing for operation in continuous space with motion constraints. We provide a rigorous analysis of the asymptotic optimality of our approach, and we present several conservative pruning strategies for modular, submodular, and time-varying information objectives. We demonstrate that our proposed approach finds optimal solutions more quickly than existing combinatorial solvers, and we provide a proof-ofconcept field implementation on an autonomous surface vehicle performing a wireless signal strength monitoring task in a lake.

High-Performance Path Planning of UAV under Complex Environment

“…The development of RRTs algorithms goes through, running the RRT multiple times, deleting and rebuilding parts of the tree and running multiple trees concurrently. Many types of research have been carried out by modifying RRTs algorithm to be more convenient for the problem of UAV path planning, Dynamic-Domain RRTs DDRRT [18], Rapidly-exploring Random Graphs (RRGs) [19], Information-rich RRT (IRRT) [20], improved RRT* algorithm which introduces D* Lite to solve path planning problems in 3-D environment for UAV is proposed in [21], adaptive RRT algorithm based on dynamic step (DRRT) for UAV path planning [22], speeding up RRTs algorithm through parallelization on large-scale distributed memory using the message passing [23], multiple trees for team of vehicles path planning [24], Closed Loop Rapidly-exploring Random Trees (CL-RRT) [25], Potential Guided Directional-RRT* [26], RRT-A* algorithm [27], Guided RRT [28], Guiding attraction based random tree (GART) [29], Medial Axis RRT (MARRT) [30], Rapidly explore random tree policy iteration (RRTPI) [31] and Variable probability based bidirectional RRT algorithm (VPB-RRT) [32].…”

Planning under uncertainty in the continuous domain: A generalized belief space approach

“…There are several complications arising when one tries to solve (11). First, in order to obtain the distribution gb (X k+l ) one has to marginalize the latent variables Γ k+1:k+l , according to (9). Second, contrarily to standard estimation problems (in which one would resort to numerical optimization techniques to solve (11)) the predicted belief gb (X k+l ) is a function of Z k+1:k+l , which are unknown at planning time.…”

“…More recently, Hollinger et. al [9] propose more efficient algorithms, based on rapidlyexploring random trees and probabilistic roadmaps. These approaches usually assume that the robot moves in a partially known environment; a remarkable property of these techniques is that they approach optimality when increasing the runtime (which is exponential in the size of the problem).…”

Towards Planning in Generalized Belief Space

“…The corresponding problem is also referred to as informative path planning. These problems are characterized by a combinatorial complexity [8], which increases with the available budget. A greedy strategy for informative path planning is proposed by Singh et al [21] while a branch and bound approach is proposed by Binney et al in [3].…”

“…More recently, Hollinger et. al [8] propose more efficient algorithms, based on rapidly-exploring random tree and probabilistic roadmap. The approaches falling in these second category usually assumes that the robot moves in a known environment; a remarkable property of these techniques is that they approach optimality when increasing the runtime (which is exponential in the size of the problem).…”