Most of the high-performance robots built to date employ hydraulic actuation. Examples such as BigDog, LS3, HyQ, and Atlas enhance the potential of this actuation technology. Hydraulics has many properties that make it an ideal choice for highly dynamic articulated robot applications:
- Higher power-to-weight ratio;
- The centralized power source (pump + tank) can provide power for several actuators. This permits to distribute better the actuation weight, keeping the robot base heavier and the links lighter;
- The hydraulic transmission, characterized by the fluid, is stiffer than electric drives transmissions (e.g. gear box), resulting in the possibility to have higher closed loop gains, greater accuracy, and better frequency response;
- Hydraulic actuators are mechanically very simple and allow for robust design against impacts and overload. This is very important in dynamic applications, where high peak forces on the robot structure cannot be avoided and are even part of the requirements
- Hydraulic actuators (e.g. valves) guarantee high enough actuator bandwidth to achieve near perfect torque control in the spectrum of interest for dynamic locomotion of medium to large scale robots. This provides legged robots, such as HyQ, the compliance needed for legged locomotion by using a naturally very stiff actuation system;
- The hydraulic fluid serves also as coolant of the actuators.
The HyQ leg prototype consists of two links and it has three actuated degrees of freedom (DOF): two in the hip (abduction/adduction and flexion/extension) and one in the knee (flexion/extension). Both flexion/extension DOF are actuated by small-sized hydraulic components: an asymmetric hydraulic cylinder (Hoerbiger LB6 1610 0080) driven by high-performance electro-hydraulic servovalves (Moog E024).
HyQ cylinder (left), valve (center), and valve assembled with a custom made manifold on the cylinder (right)
Hydraulic Control Architecture
Being able to apply precise joint torques to a robot has many advantages. Torque control allows various forms of impedance control, control of contact forces, virtual model control, as well as model-based controls, such as rigid body dynamics based control (inverse dynamics, gravity compensation), and operational space control. HyQ's legs employ a control scheme composed of two different closed-loop controls: an outer position loop and an inner torque loop. These controllers are set in cascade, where the output of the position controller, together with a feed-forward torque that comes from the rigid body inverse dynamics control, manipulates the set-point of the torque controller.
Block diagram of the cascade control employed on the HyQ's legs (Boaventura, 2012)
The key aspects for achieving high-performance torque control with a hydraulic system are:
- To use servovalves with high flow control bandwidth: HyQ uses high-spec valves, with bandwidth of about 250 Hz (for ± 25% opening), which are able to exploit the naturally high stiffness of hydraulics;
- To improve the controller using model-based control: HyQ employs model-based control approaches such as load velocity compensation and feedback linearization.
A simple model-based approach for improving torque tracking capabilities is the load velocity compensation (Boaventura, 2012). In hydraulics, to compensate for the load velocity is equivalent to feed forward an extra flow such that the pressure in the cylinder chambers is kept constant while the rod is moving. Thus, the feedback force controller has to only supply an additional small flow to adjust the pressure difference between the chambers to reach the desired force. This load motion compensation permits to significantly improve the torque response.
A more complex model-based control consists in linearizing the force dynamics by using the control approach feedback linearization (Boaventura, 2012). This approach, besides compensating for the load motion influence, also linearizes the nonlinear pressure-flow characteristics.
The very satisfactory performance obtained with the torque-controlled leg of HyQ is shown in the following video, which presents results such as:
- Friction and gravity compensation;
- Great tracking at 5 Hz sinusoidal position reference;
- Active compliance for safe human-robot interactions;
- Drop tests.