Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Interaction-aware Conformal Prediction for Crowd Navigation

Zhe Huang, Tianchen Ji, Heling Zhang, Fatemeh Cheraghi Pouria, Katherine Driggs-Campbell, Roy Dong

TL;DR

This paper tackles uncertainty in crowd navigation by addressing the feedback loop between robot actions and human motion. It introduces Interaction-aware Conformal Prediction (ICP), which interleaves trajectory prediction, conformal uncertainty quantification, and model predictive control in an iterative loop conditioned on robot plans and simulated human reactions. ICP uses a frozen GST predictor and an ORCA-based human simulator to build plan-dependent calibration data, enabling conformal prediction to yield reliable prediction intervals for human trajectories and to tighten safety margins in planning. The approach demonstrates strong performance in simulations across varying crowd densities, offering a practical balance between navigation efficiency, social awareness, and quantified uncertainty with real-time capability. These results point to meaningful real-world impact for safe, scalable robot crowd navigation and potential extensions to other interaction-rich domains.

Abstract

During crowd navigation, robot motion plan needs to consider human motion uncertainty, and the human motion uncertainty is dependent on the robot motion plan. We introduce Interaction-aware Conformal Prediction (ICP) to alternate uncertainty-aware robot motion planning and decision-dependent human motion uncertainty quantification. ICP is composed of a trajectory predictor to predict human trajectories, a model predictive controller to plan robot motion with confidence interval radii added for probabilistic safety, a human simulator to collect human trajectory calibration dataset conditioned on the planned robot motion, and a conformal prediction module to quantify trajectory prediction error on the decision-dependent calibration dataset. Crowd navigation simulation experiments show that ICP strikes a good balance of performance among navigation efficiency, social awareness, and uncertainty quantification compared to previous works. ICP generalizes well to navigation tasks under various crowd densities. The fast runtime and efficient memory usage make ICP practical for real-world applications. Code is available at https://github.com/tedhuang96/icp.

Interaction-aware Conformal Prediction for Crowd Navigation

TL;DR

This paper tackles uncertainty in crowd navigation by addressing the feedback loop between robot actions and human motion. It introduces Interaction-aware Conformal Prediction (ICP), which interleaves trajectory prediction, conformal uncertainty quantification, and model predictive control in an iterative loop conditioned on robot plans and simulated human reactions. ICP uses a frozen GST predictor and an ORCA-based human simulator to build plan-dependent calibration data, enabling conformal prediction to yield reliable prediction intervals for human trajectories and to tighten safety margins in planning. The approach demonstrates strong performance in simulations across varying crowd densities, offering a practical balance between navigation efficiency, social awareness, and quantified uncertainty with real-time capability. These results point to meaningful real-world impact for safe, scalable robot crowd navigation and potential extensions to other interaction-rich domains.

Abstract

During crowd navigation, robot motion plan needs to consider human motion uncertainty, and the human motion uncertainty is dependent on the robot motion plan. We introduce Interaction-aware Conformal Prediction (ICP) to alternate uncertainty-aware robot motion planning and decision-dependent human motion uncertainty quantification. ICP is composed of a trajectory predictor to predict human trajectories, a model predictive controller to plan robot motion with confidence interval radii added for probabilistic safety, a human simulator to collect human trajectory calibration dataset conditioned on the planned robot motion, and a conformal prediction module to quantify trajectory prediction error on the decision-dependent calibration dataset. Crowd navigation simulation experiments show that ICP strikes a good balance of performance among navigation efficiency, social awareness, and uncertainty quantification compared to previous works. ICP generalizes well to navigation tasks under various crowd densities. The fast runtime and efficient memory usage make ICP practical for real-world applications. Code is available at https://github.com/tedhuang96/icp.

Paper Structure

This paper contains 16 sections, 3 theorems, 20 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Conformal Prediction
  4. Crowd Navigation
  5. Model Predictive Control
  6. Introduction to Conformal Prediction
  7. Method
  8. Problem Formulation
  9. Preliminaries
  10. Interaction-aware Conformal Prediction
  11. Experiments
  12. Experiment Setup
  13. Experiment Results
  14. Limitations
  15. Conclusions
  16. ...and 1 more sections

Key Result

lemma thmcounterlemma

Let $X, X_1, \dots, X_n$ be exchangeable random variables. Let $X_{(k)}$ be the $k$-th smallest value among $X_1, \dots, X_n$. Then we have

Figures (4)

  • Figure 1: Interaction-aware Conformal Prediction (ICP) iteratively quantifies uncertainty of human trajectory prediction by human motion simulation under the assumption that the robot would execute the latest plan, and plans robot motion with the conformal interval radii calibrated from the latest simulation dataset.
  • Figure 2: Coverage rate (CR) and robot navigation time (NT) of algorithms with Pred-Step Execution scheme in crowd scenes of different number of humans. The error bars denote the standard deviation. The unit of robot navigation time is second. We use ICP$_1$ among all ICPs with PSE configurations for comparison.
  • Figure 3: Performance comparison between ICP and ACP-W for both Pred-Step Execution (PSE) Scheme and Single-Step Execution (SSE) Scheme. One black dot is for one test case in the Pred-Step Execution, where ICP and ACP-W share the same configurations on start and goal positions for the robot and the humans. One red triangle is for one test case in the Single-Step Execution. The X value of a black dot or a red triangle shows the performance of ICP, and the Y value shows the performance of ACP-W. Note we use ICP$_1$ for PSE comparison.
  • Figure 4: Snapshots of one crowd navigation test case in SSE scheme. We use ICP$_9$ for ICP visualization. The last column shows the snapshots whe n the robot reaches the goal. The bright yellow disk denotes the robot. The star denotes the robot goal. The orange circles with indices denote the humans with the predicted positions. The bright blue circles denote human radius bloated by the confidence interval radius. The red dots denote the history of the robot positions. The blue dots denote the generated plan to be executed by the robot. The dark yellow dots with indices denote the corresponding human's goal.

Theorems & Definitions (6)

  • lemma thmcounterlemma
  • proof
  • theorem thmcountertheorem
  • proof
  • theorem thmcountertheorem
  • proof