Quadruped Robot Mastering Obstacle Avoidance: Integrating Perception and Decision-making

Quadruped robots play a critical role in navigating complex environments, where effective perception and obstacle avoidance are paramount. Despite advancements in reinforcement learning for motion control, current research has inadequately addressed the combined challenges of traversable and non-traversable obstacles, lack of ability to combine obstacle avoidance and terrain adaptation. This paper proposes a novel motion control framework for quadruped robots that integrates deep reinforcement learning (DRL) to unify the modeling of both obstacle types, overcoming the limitations of existing methods. Our framework introduces a generalized joint obstacle avoidance strategy, which merges laser sensing and proprioception, to enhance the robot’s perception and navigation capabilities in diverse environments. Experimental results, conducted in both simulated and real-world terrains, demonstrate that our approach significantly outperforms traditional methods in obstacle avoidance performance, underscoring its potential for real-world applications.