Federated PPO: Federated Proximal Policy Optimization for Multi-Robot Collision Avoidance
Abstract
Abstract Decentralized multi-robot collision avoidance is a challenging task that requires robots to avoid both static and dynamic obstacles relying solely on their own observation. In recent years, methods based on multi-agent reinforcement learning (MARL) have been widely applied in this field, where shared policies are trained to guide the collision avoidance behaviors. However, the application of shared policies may restrict the applicability of such methods in heterogeneous multi-robot systems, while independent policies tend to exhibit lower convergence rates. In this paper, we propose the Federated Proximal Policy Optimization algorithm (FedPPO) to share experience among robots with privacy preservation, which can be applied in heterogeneous robots. The federated learning paradigm is utilized so that each robot uploads its own model parameter to the server after local training, and the server aggregates the model parameter to share experience. The experimental validation on the simulator shows that the proposed method has promising results on both homogeneous and heterogeneous multi-robot systems. FedPPO serves as an effective solution to address the low sampling efficiency of independent policy methods and the limited applicability of shared policy methods.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.