Cable-driven parallel robots (CDPRs) offer significant advantages, such as the lightweight design, large workspace, and easy reconfiguration, making them essential for various spatial applications and extreme environments. However, despite their benefits, CDPRs face challenges, notably the uncertainty in terms of the post-reconstruction parameters, complicating cable coordination and impeding mechanism parameter identification. This is especially notable in CDPRs with redundant constraints, leading to cable relaxation or breakage. To tackle this challenge, this paper introduces a novel approach using reinforcement learning to drive redundant constrained cable-driven robots with uncertain parameters. Kinematic and dynamic models are established and applied in simulations and practical experiments, creating a conducive training environment for reinforcement learning. With trained agents, the mechanism is driven across 100 randomly selected parameters, resulting in a distinct directional distribution of the trajectories. Notably, the rope tension corresponding to 98% of the trajectory points is within the specified tension range. Experiments are carried out on a physical cable-driven device utilizing trained intelligent agents. The results indicate that the rope tension remained within the specified range throughout the driving process, with the end platform successfully maneuvered in close proximity to the designated target point. The consistency between the simulation and experimental results validates the efficacy of reinforcement learning in driving unknown parameters in redundant constraint-driven robots. Furthermore, the method’s applicability extends to mechanisms with diverse configurations of redundant constraints, broadening its scope. Therefore, reinforcement learning emerges as a potent tool for acquiring motion data in cable-driven mechanisms with unknown parameters and redundant constraints, effectively aiding in the reconstruction process of such mechanisms.
