Recent interest in the control of underactuated robots has surged significantly due to the impressive athletic behaviors shown by robots developed by, e.g., Boston Dynamics ( https://www.bostondynamics.com ), Agility Robotics ( https://agilityrobotics.com/robots ), and the Massachusetts Institute of Technology [1] . This gives rise to the need for canonical robotic hardware setups for studying underactuation and comparing learning and control algorithms for their performance and robustness. Similar to OpenAIGym [2] and Stable Baselines [3] , which provide simulated benchmarking environments and baselines for reinforcement learning algorithms, there is a need for benchmarking learning and control methods on real canonical hardware setups. To encourage reproducibility in robotics and artificial intelligence research, these hardware setups should be affordable and easy to manufacture with off-the-shelf components, and the accompanying software should be open source. Acrobots and pendubots are classical textbook examples of canonical underactuated systems with strong nonlinear dynamics, and their swing-up and upright balancing is considered a challenging control problem, especially on real hardware.
