# Dynamic Channel: A Planning Framework for Crowd Navigation

arXiv: https://arxiv.org/abs/1903.00143 · HTML: https://mlsanity.com/paper/1903.00143/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Mar 2019.

## Built on by (most cited first, 10 of 10)

- [TBD Pedestrian Data Collection: Towards Rich, Portable, and Large-Scale Natural Pedestrian Data](https://mlsanity.com/paper/2309.17187/) (2023)
- [Mixed Strategy Nash Equilibrium for Crowd Navigation](https://mlsanity.com/paper/2403.01537/) (2024)
- [HeR-DRL:Heterogeneous Relational Deep Reinforcement Learning for Decentralized Multi-Robot Crowd Navigation](https://mlsanity.com/paper/2403.10083/) (2024)
- [TrajPRed: Trajectory Prediction with Region-based Relation Learning](https://mlsanity.com/paper/2404.06971/) (2024)
- [SAFER: Safe Collision Avoidance using Focused and Efficient Trajectory Search with Reinforcement Learning](https://mlsanity.com/paper/2209.11789/) (2022)
- [Urban Risk-Aware Navigation via VQA-Based Event Maps for People with Low Vision](https://mlsanity.com/paper/2605.11782/) (2026)
- [Towards Rich, Portable, and Large-Scale Pedestrian Data Collection](https://mlsanity.com/paper/2203.01974/) (2022)
- [Future-Oriented Navigation: Dynamic Obstacle Avoidance with One-Shot Energy-Based Multimodal Motion Prediction](https://mlsanity.com/paper/2505.00237/) (2025)
- [A New Trajectory-Oriented Approach to Enhancing Comprehensive Crowd Navigation Performance](https://mlsanity.com/paper/2512.06608/) (2025)
- [Learning-Based Motion Planning for Dynamic Environments: From Foundational Algorithms to Emerging Paradigms](https://mlsanity.com/paper/2608.00625/) (2026)

## Abstract (arXiv)

Real-time navigation in dense human environments is a challenging problem in robotics. Most existing path planners fail to account for the dynamics of pedestrians because introducing time as an additional dimension in search space is computationally prohibitive. Alternatively, most local motion planners only address imminent collision avoidance and fail to offer long-term optimality. In this work, we present an approach, called Dynamic Channels, to solve this global to local quandary. Our method combines the high-level topological path planning with low-level motion planning into a complete pipeline. By formulating the path planning problem as graph searching in the triangulation space, our planner is able to explicitly reason about the obstacle dynamics and capture the environmental change efficiently. We evaluate efficiency and performance of our approach on public pedestrian datasets and compare it to a state-of-the-art planning algorithm for dynamic obstacle avoidance.