Don't Freeze, Don't Crash: Extending the Safe Operating Range of Neural Navigation in Dense Crowds

Jiefu Zhang; Yang Xu; Vaneet Aggarwal

cs.LGMar 6, 2026

Don't Freeze, Don't Crash: Extending the Safe Operating Range of Neural Navigation in Dense Crowds

Authors: Jiefu Zhang, Yang Xu, Vaneet Aggarwal

Organizations: Purdue University, USA

Abstract

Navigating safely through dense crowds requires collision avoidance that generalizes beyond the densities seen during training. Learning-based crowd navigation can break under out-of-distribution crowd sizes due to density-sensitive observation normalization and social-cost scaling, while analytical solvers often remain safe but freeze in tight interactions. We propose a reinforcement learning approach for dense, variable-density navigation that attains zero-shot density generalization using a density-invariant observation encoding with density-randomized training and physics-informed proxemic reward shaping with density-adaptive scaling. The encoding represents the distance-sorted $K$ nearest pedestrians plus bounded crowd summaries, keeping input statistics stable as crowd size grows. Trained with $N\!\in\![11,16]$ pedestrians in a $3\mathrm{m}\times3\mathrm{m}$ arena and evaluated up to $N\!=\!21$ pedestrians ( $1.3\times$ denser), our policy reaches the goal in $>99\%$ of episodes and achieves $86\%$ collision-free success in random crowds, with markedly less freezing than analytical methods and a $>\!60$ -point collision-free margin over learning-based benchmark methods. Codes are available at https://github.com/jznmsl/PSS-Social.

Figures & tables

Method	Max ped/m 2	High density	Safety metric	OOD density
CADRL [ 2 ]	0.10	$\times$	✓	$\times$
SARL [ 1 ]	0.10	$\times$	✓	$\times$
DenseCAvoid [ 16 ]	$>$ 1 $\dagger$	✓ $\dagger$	$\times$	$\times$
CrowdSteer [ 17 ]	$>$ 1 $\dagger$	✓ $\dagger$	$\times$	$\times$
Frozone [ 18 ]	0.10	$\times$	✓	$\times$
DS-RNN [ 4 ]	0.40	$\times$	✓	✓

TABLE I: Comparison of DRL crowd navigation methods. Max density refers to the highest density evaluated in the original work.

Fig. 1: Overview of the PSS-Social pipeline. The simulator state is mapped to a fixed-dimensional observation consisting of ego-goal features, distance-sorted neighbor slots with $K$ -cap truncation, and bounded crowd-summary scalars. The normalized observation is passed to an MLP policy trained with PPO. The training reward adds potential-based proxemic shaping with density-adaptive scaling to the environment’s extrinsic navigation reward.

Method	$N{=}11$	$N{=}13$	$N{=}15$	$N{=}17$	$N{=}19$	$N{=}21$
PSS-Social (ours)	$\mathbf{96.2}_{\pm 2.2}$	$\mathbf{95.4}_{\pm 1.5}$	$\mathbf{94.8}_{\pm 2.9}$	$\mathbf{94.0}_{\pm 2.5}$	$\mathbf{93.6}_{\pm 2.4}$	$\mathbf{86.4}_{\pm 2.5}$
LSTM-RL	$91.2_{\pm 1.9}$	$92.8_{\pm 1.9}$	$85.8_{\pm 2.8}$	$83.6_{\pm 3.4}$	$80.8_{\pm 3.4}$	$74.8_{\pm 5.0}$
DS-RNN	$43.6_{\pm 9.0}$	$38.4_{\pm 5.5}$	$36.2_{\pm 4.5}$	$28.2_{\pm 6.8}$	$22.0_{\pm 7.8}$	$12.0_{\pm 6.5}$
SARL	$47.0_{\pm 4.3}$	$39.2_{\pm 5.3}$	$34.0_{\pm 3.7}$	$28.0_{\pm 4.4}$	$16.6_{\pm 7.2}$	$\phantom{0}9.8_{\pm 7.0}$
ORCA †	$95.2_{\pm 3.0}$	$93.8_{\pm 3.3}$	$92.8_{\pm 2.4}$	$87.8_{\pm 3.3}$	$83.6_{\pm 3.9}$	$74.8_{\pm 4.7}$
SFM †	$42.6_{\pm 5.6}$	$42.2_{\pm 1.3}$	$40.6_{\pm 4.6}$	$36.4_{\pm 4.4}$	$30.0_{\pm 4.1}$	$29.0_{\pm 3.4}$

TABLE II: Safe Success Rate (%) across the full density sweep. Mean

\pm

1 std across 5 seeds.

\dagger

: privileged oracle access.

Fig. 2: Analysis of specific failure modes across density. (a) Collisions per episode measures the frequency of physical contact, distinguishing unsafe behavior from safe navigation. (b) Freezing rate tracks the percentage of time agents spend deadlocked/stationary, identifying conservative failure modes.

Fig. 3: Episode Outcomes. Green: Safe success; Amber: Collision; Red: Timeout. Baselines shift from frequent collisions to total failure as density rises, while PSS-Social maintains robust performance.

Fig. 4: Safe Success Rate across the density sweep.

Variant	$N{=}11$	$N{=}13$	$N{=}15$	$N{=}17$	$N{=}19$	$N{=}21$
Baseline	$95.8_{\pm 1.5}$	$\mathbf{95.4}_{\pm 1.8}$	$\mathbf{95.6}_{\pm 1.7}$	$92.6_{\pm 3.0}$	$88.0_{\pm 4.1}$	$79.6_{\pm 6.7}$
PSS only	$\mathbf{97.8}_{\pm 1.6}$	$94.2_{\pm 3.3}$	$93.2_{\pm 3.0}$	$91.6_{\pm 2.7}$	$90.8_{\pm 3.2}$	$80.4_{\pm 7.4}$
PSS-Social (ours)	$96.2_{\pm 2.2}$	$95.4_{\pm 1.5}$	$94.8_{\pm 2.9}$	$\mathbf{94.0}_{\pm 2.5}$	$\mathbf{93.6}_{\pm 2.4}$	$\mathbf{86.4}_{\pm 2.5}$

TABLE III: Reward shaping ablation. Safe Success Rate (%) across the density sweep. Mean

\pm

1 std across 5 seeds.

Variant	$N{=}11$	$N{=}13$	$N{=}15$	$N{=}17$	$N{=}19$	$N{=}21$
$K$ -cap + Sort (full)	$96.2_{\pm 2.2}$	$95.4_{\pm 1.5}$	$94.8_{\pm 2.9}$	$94.0_{\pm 2.5}$	$93.6_{\pm 2.4}$	$86.4_{\pm 2.5}$
No $K$ -cap + Sort	$96.2_{\pm 2.2}$	$95.4_{\pm 1.5}$	$94.8_{\pm 2.9}$	$25.4_{\pm 20.8}$	$5.6_{\pm 3.4}$	$2.0_{\pm 1.6}$
$K$ -cap + No Sort	$39.0_{\pm 5.4}$	$32.0_{\pm 2.4}$	$27.0_{\pm 5.8}$	$24.4_{\pm 2.9}$	$19.2_{\pm 6.3}$	$14.8_{\pm 2.5}$
No $K$ -cap + No Sort	$39.0_{\pm 5.4}$	$32.0_{\pm 2.4}$	$27.0_{\pm 5.8}$	$7.0_{\pm 4.7}$	$2.8_{\pm 2.2}$	$1.0_{\pm 1.4}$

TABLE IV: Observation encoding ablation. Safe Success Rate (%) across the density sweep. Mean

\pm

1 std across 5 seeds.

Safe Robot Navigation

Safe robot navigation develops methods for reaching goals while avoiding collisions and managing uncertainty, dynamic obstacles, and environmental hazards.

See all 138 papers

Diffusion Policies for Short-Horizon Planning in Robot Crowd Navigation
Aug 27, 2026Wendong Li, Jochen GarckeRobot Navigation Robotic RL
Institute for Numerical Simulation University of Bonn Bonn, Germany
Navigating the Proximity-Safety Balance: Constraint Decomposition for Human Following in Pedestrian Crowds
Aug 10, 2026Shiting Gong, Jianpeng Yao, Jinfeng Wang +2Constrained RL Robotic RL
University of Pennsylvania, PA, USA · University of California, Riverside, CA, USA · 3Stanford University, CA, USA +2
Risk-Sensitive Crowd Navigation with Adaptive Ellipsoidal Conformal Prediction
Oct 5, 2026Ruihan A. Li, Ziyao Guo, Yingying LiDistribution Shift Robustness Robot Navigation
Coordinated Science Laboratory, University of Illinois at Urbana-Champaign, Urbana, IL

Aug 27, 2026J/K move · Enter open · S save

Diffusion Policies for Short-Horizon Planning in Robot Crowd Navigation

Wendong Li, Jochen Garcke

Institute for Numerical Simulation University of Bonn Bonn, Germany

Robot Navigation Robotic RL Diffusion-Based Planning Diffusion Policy Action Chunking

Open paper AskPDF arXiv

Robot crowd navigation requires safe and efficient decision-making under dense, dynamic, and multimodal human--robot interactions. Existing reinforcement-learning methods typically output a single reactive action at each timestep, which limits their ability to represent diverse short-term avoidance strategies. We propose Planning Diffusion Policy Optimization (PDPO), an offline-to-online reinforcement-learning framework that uses a diffusion policy to generate short-horizon action chunks for crowd navigation. PDPO is first pretrained on collision-avoidance demonstrations and then fine-tuned online with PPO by treating the denoising process as an internal decision process. During execution, the policy generates a five-step action chunk and applies it in a receding-horizon manner. Furthermore, we observe an evaluation artifact in common crowd-navigation benchmarks: without explicit boundary constraints, learned agents may leave the valid domain and bypass dense crowds. To address this, we introduce a setting in which boundary violations are treated as collisions. Experiments show that PDPO obtains an improved success rate over strong baselines, and ablations demonstrate that action chunks are especially important for the modified bounded benchmark.

Abstract

Safe Robot Navigation

Explore similar work

Diffusion Policies for Short-Horizon Planning in Robot Crowd Navigation

Navigating the Proximity-Safety Balance: Constraint Decomposition for Human Following in Pedestrian Crowds

Risk-Sensitive Crowd Navigation with Adaptive Ellipsoidal Conformal Prediction