Abstract:We present a novel learning-based collision avoidance algorithm, CrowdSteer, for mobile robots operating in dense and crowded environments. Our approach is end-to-end and uses multiple perception sensors such as a 2-D lidar along with a depth camera to sense surrounding dynamic agents and compute collision-free velocities. Our training approach is based on the sim-to-real paradigm and uses high fidelity 3-D simulations of pedestrians and the environment to train a policy using Proximal Policy Optimization (PPO). We show that our learned navigation model is directly transferable to previously unseen virtual and dense real-world environments. We have integrated our algorithm with differential drive robots and evaluated its performance in narrow scenarios such as dense crowds, narrow corridors, T-junctions, L-junctions, etc. In practice, our approach can perform real-time collision avoidance and generate smooth trajectories in such complex scenarios. We also compare the performance with prior methods based on metrics such as trajectory length, mean time to goal, success rate, and smoothness and observe considerable improvement.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how mobile robots can achieve real - time obstacle avoidance in a dense crowd environment. Specifically, the paper proposes a new learning - based obstacle avoidance algorithm, CrowdSteer, which aims to enable mobile robots to navigate reliably in high - density crowds and avoid collisions with obstacles or pedestrians. The paper focuses particularly on using multi - sensor fusion techniques to improve the obstacle avoidance performance of robots in complex environments, especially in scenarios such as narrow corridors, T - junctions, and L - junctions, where robots need to make real - time decisions to generate smooth obstacle - avoidance trajectories.
### Main Problems
1. **Real - time Obstacle Avoidance in High - Density Crowds**: In environments with a relatively high population density, such as 1 - 3 people per square meter or more, the motion trajectories of pedestrians are usually not smooth and may change suddenly, which makes it difficult to predict their motion trajectories. In addition, it is difficult to predict the behavior of pedestrians due to occlusion or non - smooth motion.
2. **Multi - sensor Fusion**: A single sensor performs poorly in dense or cluttered scenes, so it is necessary to combine multiple sensing sensors (such as 2D LiDAR and depth cameras) to improve obstacle avoidance performance.
3. **Transfer from Simulation to the Real Environment**: How to train a model through high - fidelity 3D simulation so that it can operate effectively in the real world and reduce the gap between simulation and reality.
### Solutions
1. **CrowdSteer Algorithm**: This algorithm uses deep reinforcement learning (specifically Proximal Policy Optimization, PPO) to train obstacle - avoidance strategies. Through multi - sensor fusion, the algorithm can perceive the surrounding environment more accurately and generate smooth obstacle - avoidance trajectories.
2. **Multi - sensor Fusion**: By combining the data of 2D LiDAR and depth cameras, the algorithm can better identify the characteristics of dynamic obstacles and pedestrians, thereby making more reasonable obstacle - avoidance decisions.
3. **High - Fidelity Simulation**: Using a high - fidelity 3D simulation environment for training, these environments contain real pedestrian motion and occlusion scenarios, which helps to improve the generalization ability of the model in the actual environment.
### Experimental Results
- **Performance Evaluation**: The paper tested the algorithm in different indoor environments, including narrow corridors, T - junctions, and L - junctions. The experimental results show that CrowdSteer is superior to existing traditional methods (such as Dynamic Window Approach, DWA) and the state - of - the - art learning methods in terms of success rate, time to reach the target, and trajectory length.
- **Comparative Experiment**: Compared with Fan et al.'s method, CrowdSteer reduces the time by 68.16% and the trajectory length by 6.12%.
### Conclusion
The CrowdSteer algorithm successfully solves the problem of real - time obstacle avoidance of mobile robots in high - density crowd environments through multi - sensor fusion and deep reinforcement learning. This algorithm not only performs well in the simulation environment but also has good generalization ability and robustness in practical applications.