What an electro-hydrostatic actuator is

A classic hydraulic humanoid, like the hydraulic Atlas that Boston Dynamics retired in 2024, works like a small excavator. One pump keeps a supply line at high pressure. At every joint, a servo valve meters that pressure into a cylinder. The valve is what controls the joint, and it does so by throttling: it lets through only part of the available pressure and turns the rest into heat.

An EHA moves the power source into the joint. A brushless motor turns a small reversible pump. The pump’s two ports connect straight to the two chambers of a double-acting cylinder, or to a rotary vane motor. Turn the motor one way and oil flows into one chamber and out of the other, and the piston moves. Reverse the motor and the piston reverses. The oil never leaves the loop, so there are no supply hoses running through the body. Only electrical power and data reach the joint, as in an electric humanoid.

How the pump replaces the gearbox

An electric joint needs a reducer because a motor is fast and weak. An EHA uses the oil as its reducer. Each pump revolution moves a fixed volume of oil, its displacement. The motor torque needed scales with pressure times displacement, while the piston force is pressure times piston area. A pump with a tiny displacement driving a much larger piston gives a large effective ratio, set by geometry rather than by gear teeth. Kaminaga’s 2009 knee EHA reached a reduction of about 100:1 this way.

The same relation makes force easy to measure. Pressure sensors on the two lines give the piston force directly, which lets an EHA close a fast inner pressure loop without a separate torque sensor. Hydra’s linear EHA delivers 1,500 N at 5.25 MPa of differential pressure, which works out to about 2.9 cm² of effective piston area. Its maximum piston speed is about 150 mm/s.

Why drop the servo valve

Servo-valve hydraulics give the highest torque density of any actuator type, and they gave Atlas its jumps and backflips. The cost is energy. The central pump must hold the whole system at supply pressure whether or not a joint needs it, and every valve throttles the difference. A 2023 paper from the Beijing Institute of Technology puts valve-controlled cylinder transmission efficiency below 30%. It claims an EHA can exceed 70% because there is no throttling loss. Treat both as design-level figures, not a side-by-side measurement.

Boston Dynamics worked around the problem with custom valves. Speaking at the 2018 IFK fluid power conference, engineering VP Aaron Saunders described valves with a braking mode for negative work and a coasting mode that moves oil chamber to chamber. Each of the 80 kg robot’s 28 degrees of freedom ran force control. The EHA takes the opposite route: it removes the shared high-pressure supply entirely.

Hydra: a whole humanoid on EHAs

The most complete EHA humanoid is Hydra, built by Yoshihiko Nakamura’s group at the University of Tokyo and shown at the ISER 2016 symposium. It is 1.80 m tall, weighs 130 kg and has 40 joints. They are driven by 24 double-rod linear cylinders and 8 vane motors, fed by 36 water-cooled trochoid pumps. Each hand uses a five-DoF cluster EHA. Multi-axis joints use parallel mechanisms so that several cylinders share the load.

Tianyi Ko’s 2018 thesis documents the result. Joint position and torque control reached roughly 30 to 40 Hz bandwidth. Hydra walked with a 0.2 m stride and 1 s steps, and it absorbed pushes on its limbs through compliant whole-body control. Its pressure loop reached 100 Hz with the piston held fixed, which the thesis reports as higher than earlier EHA and series elastic actuator results.

Back-drivability: the knee test

The Tokyo group’s motive was back-drivability: a joint that yields when pushed, so a robot can absorb contact. Oil flowing through a pump has little friction compared with a high-ratio gear train. In 2009, Kaminaga and colleagues compared a knee EHA with a Harmonic Drive CSG40-100 of almost the same size. The EHA was 130 mm across and weighed 2.0 kg, with 1,988 N·m allowable torque. The strain-wave gear was 135 mm across, weighed 1.7 kg and was rated at 2,700 N·m.

On back-driving torque, the result was mixed. The EHA’s output side measured 4 N·m, close to its 3.7 N·m datasheet estimate. The total measured figure was 14 N·m, which the authors put down mainly to oil-seal friction, against about 7.4 N·m for the gear with an assumed motor friction. The authors expected a design change to halve that figure, and aimed for a 50 to 100 times reduction long term. Ko’s later actuators attacked seal friction with thin double rods and active friction compensation.

Published EHAs for legged robots

Outside Tokyo, EHA work has moved toward heavier machines and new layouts. Pressures range from under 1 MPa in Sherbrooke’s low-pressure legs to 30 MPa in a wheel-legged knee. That spread matters, because pressure decides how small the cylinder can be and how hard the seals work.

The Beijing Institute of Technology knee EHA shows how high pressure buys force density. At its 30 MPa rating it produced 9,401 N from 3.69 kg, a force-to-weight ratio above 2,518 N/kg, against the 1,424 N/kg the same paper lists for Hydra’s actuator. In Canada, the Université de Sherbrooke takes the opposite approach: low pressure, and motorised valves that let one motor serve both legs. In France, KALYSTA Actuation, spun out of Université Paris-Saclay’s IBISC lab, is building a decentralised electrohydraulic humanoid and targets a full demonstrator in 2029.

The catch

Shrinking an EHA to humanoid scale works against it. Ko’s thesis found that miniaturisation raises internal leakage and cuts output force. Its remedies were tighter gaps, more viscous fluid, reinforced casings and water-cooled pump housings. Seals add friction that the gearless design was meant to remove. Every joint carries its own pump, oil volume and pressure sensors, and the oil adds weight on top of the motor. Hydra’s linear EHA peaked at about 40% efficiency, well short of the 70% cited as an EHA’s potential.

Commercial humanoid makers have gone the other way. Electric joints with strain-wave, planetary or roller-screw transmissions now dominate production robots, and the electric Atlas replaced the hydraulic one. We found no humanoid in commercial production that uses EHA joints. A 2025 review in China’s Journal of Mechanical Engineering still names legged robots as a key EHA application. Hydraulics suppliers such as Domin also pitch EHAs for hips, knees and shoulders, where the loads are highest.

• Watch force-to-weight at humanoid scale. The 2,500 N/kg figures come from heavy wheel-legged robots. Ask what an EHA weighs at the force a 60 kg humanoid’s knee needs.
• Ask for measured efficiency. Claims of 70% and more are common. Published humanoid-scale measurements are closer to 40% at peak.
• Ask about seals and leakage over life. Seal friction set the back-drive torque in the Tokyo knee, and seals wear.

FAQ

Sources