Joint Multi-Policy Behavior Estimation and Receding-Horizon Trajectory Planning for Automated Urban Driving

Abstract

When driving in urban environments, an autonomous vehicle must account for the interaction with other traffic participants. It must reason about their future behavior, how its actions affect their future behavior, and potentially consider multiple motion hypothesis. In this paper we introduce a method for joint behavior estimation and trajectory planning that models interaction and multi-policy decisionmaking. The method leverages Partially Observable Markov Decision Processes to estimate the behavior of other traffic participants given the planned trajectory for the ego-vehicle, and Receding-Horizon Control for generating safe trajectories for the ego-vehicle. To achieve safe navigation we introduce chance constraints over multiple motion policies in the recedinghorizon planner. These constraints account for uncertainty over the behavior of other traffic participants. The method is capable of running in real-time and we show its performance and good scalability in simulated multi-vehicle intersection scenarios.

A Review of Tracking and Trajectory Prediction Methods for Autonomous Driving

“…The authors of [128] use Partially Observable Markov Decision Processes (POMDPs) for behavior prediction and nonlinear receding horizon control, or model predictive control, for trajectory planning. The POMDP models the interactions between the ego vehicle and the obstacles.…”

Learning Interaction-aware Guidance Policies for Motion Planning in Dense Traffic Scenarios

“…Yet, solving a POMDP online can become infeasible if the right assumptions on the state, action and observation space are not made. For instance, [25] proposed to use Monte Carlo Tree Search (MCTS) algorithms to obtain an approximate optimal solution online and [26] improved the interaction modeling by proposing to feedback the vehicle commands into planning. These methods demonstrated promising results but are limited to environments for which they were specifically designed, demand high computational power and can only consider a discrete set of actions.…”

Contingencies from Observations: Tractable Contingency Planning with Learned Behavior Models

“…Hardy and Campbell [11] Passive Bandyopadhyay et al [20] Active Xu et al [21] None Zhan et al [12] Passive Sadigh et al [15] None Galceran et al [13] Active Schmerling et al [16] None Rhinehart et al [5] None Zhou et al [22] None Fisac et al [14] Active Zeng et al [6] None Tang and Salakhutdinov [17] None Cui et al [23] Passive Bajcsy et al [24] Passive…”