Sparse interacting Gaussian processes: Efficiency and optimality theorems of autonomous crowd navigation

Abstract

We study the sparsity and optimality properties of crowd navigation and find that existing techniques do not satisfy both criteria simultaneously: either they achieve optimality with a prohibitive number of samples or tractability assumptions make them fragile to catastrophe. For example, if the human and robot are modeled independently, then tractability is attained but the planner is prone to overcautious or overaggressive behavior. For sampling based motion planning of joint human-robot cost functions, for nt agents and T step lookahead, O(2 2n t T ) samples are needed for coverage of the action space. Advanced approaches statically partition the action space into free-space and then sample in those convex regions. However, if the human is moving into free-space, then the partition is misleading and sampling is unsafe: free space will soon be occupied. We diagnose the cause of these deficiencies-optimization happens over trajectory space-and propose a novel solution: optimize over trajectory distribution space by using a Gaussian process (GP) basis. We exploit the "kernel trick" of GPs, where a continuum of trajectories are captured with a mean and covariance function. By using the mean and covariance as proxies for a trajectory family we reason about collective trajectory behavior without resorting to sampling. The GP basis is sparse and optimal with respect to collision avoidance and robot and crowd intention and flexibility. GP sparsity leans heavily on the insight that joint action space decomposes into free regions; however, the decomposition contains feasible solutions only if the partition is dynamically generated. We call our approach O(2 n t )-sparse interacting Gaussian processes.

Human-Like Motion Planning Based on Game Theoretic Decision Making

“…2. Highly interesting results are published in [14,38,48,77]. Apart from examining this, future work will mainly concentrate on further experimental studies with a real robotic platform, as in Fig.…”

SocNavBench: A Grounded Simulation Testing Framework for Evaluating Social Navigation

“…More recently, topology concepts have been employed in modeling how pedestrians unanimously reach a common meta-level passing strategy in a game theoretic setting [34]. Trautman et al attempted to model interaction via a joint density term inside the pedestrians' Gaussian process mixtures models [51]. Other works explicitly modeled certain aspects of interaction and incorporated them in navigation, such as grouping considerations [22,38,62], proxemics [6,55] and personality traits [6].…”

Model Predictive Contouring Control for Collision Avoidance in Unstructured Dynamic Environments

“…The previous approaches do not account for the interaction effects between the agents and may fail in crowded scenarios, a problem known as the freezing robot problem (FRP) [17]. Interaction Gaussian Processes (IGP) [18] can be used to model each individual's path. The interactions are modeled with a nonlinear potential function.…”