Robotic locomotion in unstructured terrain demands an architecture that is at once agile, adaptive, and energy-efficient. Conventional legged robots rely on rigid electromagnetic motors and stiff transmissions, which struggle to match the grace and economy of animal movement. Drawing inspiration from biological musculoskeletal systems, we present a robotic leg actuated by antagonistic pairs of electrohydraulic artificial muscles (Peano-HASEL actuators). Like their biological counterparts, these soft actuators contract and relax in opposition to drive the joints, providing intrinsically tunable joint stiffness and compliance.
The resulting leg performs powerful and agile gait motions beyond 5 Hz and high jumps reaching up to 40% of the leg height. Its inherent compliance and tunable stiffness let it adapt to obstacles and absorb impacts passively, so it can hop over grass, sand, gravel, pebbles, and large rocks using only open-loop force control—without external terrain sensing. Capacitive self-sensing of the artificial muscles further enables proprioception and obstacle detection from the actuators' own electrical signals. Crucially, the electrohydraulic drive consumes only about 1.2% of the energy that an equivalent electromagnetic actuation system would require for the same task. This musculoskeletal architecture charts a path toward agile, robust, and energy-efficient legged robots for real-world, unstructured environments.
>5 Hz
Agile gait motion frequency
40%
Jump height relative to leg height (~128 mm)
1.2%
Energy use vs. an electromagnetic leg
5 terrains
Grass, sand, gravel, pebbles & rocks
Each muscle is a Peano-HASEL (Hydraulically Amplified Self-healing ELectrostatic) actuator: a liquid-filled plastic pouch with electrode regions on its surface. When a high voltage is applied, electrostatic (Maxwell) forces pull the electrodes together, displacing the internal liquid and causing the pouch to contract—just as a biological muscle shortens. Removing the voltage lets it relax. These actuators are lightweight, soft, fast, and energy-efficient.
Muscles can only pull, not push—so, as in vertebrates, they are arranged in antagonistic pairs (an "extensor" and a "flexor") across each joint. Activating one side moves the joint in one direction; activating the other reverses it. Co-activating both pulls the joint stiff, while relaxing both makes it compliant. This gives the leg tunable joint stiffness, the same trick animals use to stay both powerful and gentle.
The intrinsic compliance lets the leg adapt to uneven ground and absorb impacts passively, hopping over grass, sand, gravel, pebbles, and rocks under simple open-loop force control. Because electrohydraulic actuation has no continuous resistive losses like a stalled electric motor, the leg uses a small fraction of the energy a comparable electromagnetic system would need.
The same actuators that move the leg also sense: capacitive self-sensing reads back joint state and detects obstacles directly from the muscles' voltage and current, without dedicated external sensors. On the control side, a multithreaded C++ pipeline runs a cascaded task-space controller at up to 500 Hz, driving four muscle pairs through high-voltage amplifiers to track leg-tip trajectories (ellipse, rectangle, infinity, and star paths).
@article{buchner2024electrohydraulic,
title = {Electrohydraulic musculoskeletal robotic leg for agile, adaptive, yet energy-efficient locomotion},
author = {Buchner, Thomas J. K. and Fukushima, Toshihiko and Kazemipour, Amirhossein
and Gravert, Stephan-Daniel and Prairie, Manon and Romanescu, Pascal
and Arm, Philip and Zhang, Yu and Wang, Xingrui and Zhang, Steven L.
and Walter, Johannes and Keplinger, Christoph and Katzschmann, Robert K.},
journal = {Nature Communications},
volume = {15},
number = {1},
pages = {7634},
year = {2024},
doi = {10.1038/s41467-024-51568-3},
url = {https://www.nature.com/articles/s41467-024-51568-3},
publisher = {Nature Publishing Group}
}