CrowdSurfer: Sampling Optimization Augmented with Vector-Quantized Variational AutoEncoder for Dense Crowd Navigation

Naman Kumar*1, Antareep Singha*1, Laksh Nanwani*1, Dhruv Potdar1, Tarun R.1, Fatemeh Rastgar2, Simon Idoko2, Arun Kumar Singh2, K. Madhava Krishna1
1Robotics Research Center, IIIT Hyderabad, India; 2University of Tartu, Estonia
*These authors contributed equally to this work
Paper arXiv Video Code
Note - Submitted to IEEE ICRA 2025
CrowdSurfer Teaser

CrowdSurfer in Action

Abstract

Navigation amongst densely packed crowds remains a challenge for mobile robots. The complexity increases further if the environment layout changes making the prior computed global plan infeasible. In this paper, we show that it is possible to dramatically enhance crowd navigation by just improving the local planner. Our approach combines generative modelling with inference time optimization to generate sophisticated long-horizon local plans at interactive rates. More specifically, we train a Vector Quantized Variational AutoEncoder to learn a prior over the expert trajectory distribution conditioned on the perception input. At run-time, this is used as an initialization for a sampling-based optimizer for further refinement. Our approach does not require any sophisticated prediction of dynamic obstacles and yet provides state-of-the- art performance. In particular, we compare against the recent DRL-VO approach and show a 40% improvement in success rate and a 6% improvement in travel time.

Wheelchair Test - Indoor Cluttered Environment - with Globan Plan

Husky Test - Outdoors with Dynamic Obstacles - No Global Plan

Simulation Tests - Extremely Cluttered Environments

BibTeX

@misc{kumar2024crowdsurfersamplingoptimizationaugmented,
                title={CrowdSurfer: Sampling Optimization Augmented with Vector-Quantized Variational AutoEncoder for Dense Crowd Navigation}, 
                author={Naman Kumar and Antareep Singha and Laksh Nanwani and Dhruv Potdar and Tarun R and Fatemeh Rastgar and Simon Idoko and Arun Kumar Singh and K. Madhava Krishna},
                year={2024},
                eprint={2409.16011},
                archivePrefix={arXiv},
                primaryClass={cs.RO},
                url={https://arxiv.org/abs/2409.16011}, 
          }