Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-02-0193
Inverse Dynamics of Cable-driven 2D Serial Linkage
Department of Mechanical Engineering, Seoul National University, Gwanak-Gu, Seoul, 08826, South Korea
ABSTRACT
Cable-driven mechanism has been consistently attracted attention by its high payload-weight ratio, effective space utilization, and precise controllability. However, cable routing designs for precise control often sacrifice dexterity with precision, in order to fit theoretical models, which makes it difficult to take full advantage of cable driven robots. This study attempts to approach from the new perspective. This study targets serial linkages where each cable is complexly connected. A 2D hyper-redundant cable-driven linkage system was defined as an environment, and the reinforcement learning agent controlled its endpoint without having full information of the environment.
Linkage system includes 3 links (ground link, link1, link2), 2 joints (fixed joint, movable joint), 14 cables, and 8 routers. Each link is defined by length, mass, center of gravity, and thickness. Joints are modeled as damper, and have a limited range of motion. Cables are attached on links and control links by exerting force in the direction of its tension. Cables have limited range of tension, and the cable breaks when the tension exceeding the limit is exerted. Routers are intermediate points through which cables are directed. Cables can be directed through multiple links through routers, which makes a cable to affect dynamics of multiple links. This study aims to solve the endpoint control of the extremely redundant mass spring damper system without information about the internal structure. The defined system is hyper-redundant and complexly routed to mimic human tendon system. In addition, compared to conventional mass spring damper system control, this approach is able to cope with systems with large number of inputs and stochastic outputs. This study is also expected to be applicable to optimizing number and routing of cable-driven robots.
Keywords: Inverse dynamics, Cable-driven robot, Manipulation and control, Reinforcement learning.
Department of Mechanical Engineering, Seoul National University, Gwanak-Gu, Seoul, 08826, South Korea
ABSTRACT
Cable-driven mechanism has been consistently attracted attention by its high payload-weight ratio, effective space utilization, and precise controllability. However, cable routing designs for precise control often sacrifice dexterity with precision, in order to fit theoretical models, which makes it difficult to take full advantage of cable driven robots. This study attempts to approach from the new perspective. This study targets serial linkages where each cable is complexly connected. A 2D hyper-redundant cable-driven linkage system was defined as an environment, and the reinforcement learning agent controlled its endpoint without having full information of the environment.
Linkage system includes 3 links (ground link, link1, link2), 2 joints (fixed joint, movable joint), 14 cables, and 8 routers. Each link is defined by length, mass, center of gravity, and thickness. Joints are modeled as damper, and have a limited range of motion. Cables are attached on links and control links by exerting force in the direction of its tension. Cables have limited range of tension, and the cable breaks when the tension exceeding the limit is exerted. Routers are intermediate points through which cables are directed. Cables can be directed through multiple links through routers, which makes a cable to affect dynamics of multiple links. This study aims to solve the endpoint control of the extremely redundant mass spring damper system without information about the internal structure. The defined system is hyper-redundant and complexly routed to mimic human tendon system. In addition, compared to conventional mass spring damper system control, this approach is able to cope with systems with large number of inputs and stochastic outputs. This study is also expected to be applicable to optimizing number and routing of cable-driven robots.
Keywords: Inverse dynamics, Cable-driven robot, Manipulation and control, Reinforcement learning.
