Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Motion Planning under Uncertainty: Integrating Learning-Based Multi-Modal Predictors into Branch Model Predictive Control

Mohamed-Khalil Bouzidi, Bojan Derajic, Daniel Goehring, Joerg Reichardt

TL;DR

The paper addresses motion planning for autonomous vehicles under multi-modal uncertainty in other agents' behavior. It integrates a learning-based multi-modal predictor (Motion Transformer) into Branch Model Predictive Contouring Control, augmented by a topology-based scenario selection and an adaptive decision-postponing mechanism to determine the branching time. Key contributions include a scalable scenario-selection pipeline using Uniform Visibility Deformation clustering and collision-risk ranking, plus an online method to postpone commitment until uncertainty is reduced. Evaluations in traffic intersections and highway merging demonstrate improved comfort and safety with real-time performance, outperforming several baselines and fixed-branching strategies.

Abstract

In complex traffic environments, autonomous vehicles face multi-modal uncertainty about other agents' future behavior. To address this, recent advancements in learningbased motion predictors output multi-modal predictions. We present our novel framework that leverages Branch Model Predictive Control(BMPC) to account for these predictions. The framework includes an online scenario-selection process guided by topology and collision risk criteria. This efficiently selects a minimal set of predictions, rendering the BMPC realtime capable. Additionally, we introduce an adaptive decision postponing strategy that delays the planner's commitment to a single scenario until the uncertainty is resolved. Our comprehensive evaluations in traffic intersection and random highway merging scenarios demonstrate enhanced comfort and safety through our method.

Motion Planning under Uncertainty: Integrating Learning-Based Multi-Modal Predictors into Branch Model Predictive Control

TL;DR

The paper addresses motion planning for autonomous vehicles under multi-modal uncertainty in other agents' behavior. It integrates a learning-based multi-modal predictor (Motion Transformer) into Branch Model Predictive Contouring Control, augmented by a topology-based scenario selection and an adaptive decision-postponing mechanism to determine the branching time. Key contributions include a scalable scenario-selection pipeline using Uniform Visibility Deformation clustering and collision-risk ranking, plus an online method to postpone commitment until uncertainty is reduced. Evaluations in traffic intersections and highway merging demonstrate improved comfort and safety with real-time performance, outperforming several baselines and fixed-branching strategies.

Abstract

In complex traffic environments, autonomous vehicles face multi-modal uncertainty about other agents' future behavior. To address this, recent advancements in learningbased motion predictors output multi-modal predictions. We present our novel framework that leverages Branch Model Predictive Control(BMPC) to account for these predictions. The framework includes an online scenario-selection process guided by topology and collision risk criteria. This efficiently selects a minimal set of predictions, rendering the BMPC realtime capable. Additionally, we introduce an adaptive decision postponing strategy that delays the planner's commitment to a single scenario until the uncertainty is resolved. Our comprehensive evaluations in traffic intersection and random highway merging scenarios demonstrate enhanced comfort and safety through our method.
Paper Structure (13 sections, 11 equations, 7 figures, 1 table)

This paper contains 13 sections, 11 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Learning-based Multi-Modal Motion Predictor
  4. Nominal Model Predictive Contouring Control
  5. Branch Model Predictive Contouring Control
  6. Scenario Selection
  7. Topology-based Clustering
  8. Ranking and Selection based on Collision Risk
  9. Adaptive Decision Postponing
  10. Performance Evaluation
  11. Traffic Intersection
  12. Highway Merging
  13. Conclusion

Figures (7)

  • Figure 1: An illustration of the challenge of planning in traffic scenes with uncertain multi-modal predictions. To address this, Branch MPC optimizes over a scenario tree.
  • Figure 2: Our learning-aided Motion Planning and Control Framework using a Multi-modal Predictor, Scenario Selection, Adaptive Decision Postponing and Branch MPCC
  • Figure 3: Illustration of the Scenario Selection procedure. Predictions enclosed in grey ellipses belong to same cluster. Red ellipses mark scenarios selected from each cluster.
  • Figure 4: Procedure of the online estimation of the branching time based on the Bhattacharya Distance
  • Figure 5: Prediction of TP (blue car) and the plan of the AV (yellow car) at a certain timestep. Predictor outputs 6 predictions (shown in different colors) from which 2 most-relevant are selected. Corresponding trajectories are shown in corresponding colors. Right: Planned velocity profile.
  • ...and 2 more figures

Theorems & Definitions (4)

  • Remark 1
  • Definition 1
  • Definition 2
  • Definition 3