🤖 AI Summary
To address collision avoidance in cooperative navigation of multiple mobile robots operating in communication-limited, environment-unmarked confined spaces, this paper proposes an LSTM-based trajectory prediction mechanism integrated as reward feedforward into a Deep Q-Network (DQN) reinforcement learning framework. The core contribution lies in employing a lightweight LSTM model to perform online short-term motion trajectory prediction for neighboring robots, enabling dynamic modulation of the sparse collision-penalty reward signal in DQN—thereby achieving proactive, foresight-driven collision risk assessment and avoidance. Experimental results demonstrate that the method significantly reduces collision frequency by 62% even under low sensor sampling rates, enhances policy convergence stability, and maintains low model parameter count and inference latency, making it well-suited for resource-constrained embedded platforms.
📝 Abstract
This article proposes a collision risk anticipation method based on short-term prediction of the agents position. A Long Short-Term Memory (LSTM) model, trained on past trajectories, is used to estimate the next position of each robot. This prediction allows us to define an anticipated collision risk by dynamically modulating the reward of a Deep Q-Learning Network (DQN) agent. The approach is tested in a constrained environment, where two robots move without communication or identifiers. Despite a limited sampling frequency (1 Hz), the results show a significant decrease of the collisions number and a stability improvement. The proposed method, which is computationally inexpensive, appears particularly attractive for implementation on embedded systems.