In the realm of energy and robotics, a team of researchers from the University of Western Australia has developed a novel approach to enhance the safety and efficiency of human-robot collaboration. The team, led by Ning Liu and including Sen Shen, Zheng Li, Matthew D’Souza, Jen Jen Chung, and Thomas Braunl, has introduced a hybrid learning-control framework called Adaptive Reinforcement and Model Predictive Control Switching (ARMS). This research was recently published in a prominent robotics and automation journal.
The challenge addressed by the researchers is the navigation of mobile collaborative robots under the guidance of humans, while ensuring safety and adhering to proximity regulations. The ARMS framework integrates two key components: a reinforcement learning follower trained with Proximal Policy Optimization (PPO) and an analytical one-step Model Predictive Control (MPC) formulated as a quadratic program safety filter. This hybrid approach aims to leverage the strengths of both methods, ensuring safe and efficient navigation in complex environments.
To achieve robust perception under partial observability and non-stationary human motion, ARMS employs a decoupled sensing architecture. This includes a Long Short-Term Memory (LSTM) temporal encoder for the human-robot relative state and a spatial encoder for 360-degree LiDAR scans. The core innovation of ARMS is a learned adaptive neural switcher that performs context-aware soft action fusion between the two controllers. This switcher favors conservative, constraint-aware QP-based control in low-risk regions and progressively shifts control authority to the learned follower in highly cluttered or constrained scenarios where maneuverability is critical.
The researchers conducted extensive evaluations of ARMS against other navigation methods, including Pure Pursuit, Dynamic Window Approach (DWA), and an RL-only baseline. The results demonstrated that ARMS achieved an 82.5 percent success rate in highly cluttered environments, outperforming DWA and RL-only approaches by 7.1 percent and 3.1 percent, respectively. Additionally, ARMS reduced average computational latency by 33 percent to 5.2 milliseconds compared to a multi-step MPC baseline. Simulation transfers in Gazebo and initial real-world deployments further indicated the practicality and robustness of ARMS for safe and efficient human-robot collaboration.
For the energy sector, this research holds significant potential for improving the safety and efficiency of robotic operations in various environments. For instance, in the maintenance and inspection of energy infrastructure, such as wind turbines or pipelines, human-robot collaboration is crucial. The ARMS framework could enhance the navigation capabilities of robots, ensuring they operate safely and effectively in close proximity to human workers. This could lead to reduced downtime, improved safety, and increased productivity in the energy industry.
The source code and a demonstration video of the ARMS framework are available on the project’s GitHub page, providing further insights into the practical applications and robustness of this innovative approach.
This article is based on research available at arXiv.