Robot control

Virtual constraint

Definition

A virtual constraint is a relation among a robot's configuration variables that is imposed by feedback instead of by a physical linkage. In legged locomotion, virtual constraints often coordinate joint trajectories around a state-based gait phase.

Also known as: Virtual constraints

Updated

Feedback replaces a mechanical linkage

A physical constraint ties coordinates together through geometry, such as a rigid linkage. A virtual constraint specifies a desired relation and uses actuators plus feedback to enforce it. For a walking robot, an output might be the difference between an actual joint angle and a desired angle indexed by gait phase. Driving that output to zero coordinates several joints without adding hardware.

Grizzle and colleagues define virtual constraints as state-variable relations imposed through time-invariant feedback. A state-based phase variable can make the relation advance with the robot's configuration rather than with a wall clock, which helps coordinate a gait when timing varies.

Use in biped control

Virtual constraints can specify torso posture, swing-foot clearance, leg shape, or other outputs. They are central to one construction of hybrid zero dynamics, where the controlled relations also need to remain consistent through foot impact.

They can be combined with other control layers. Gong and colleagues report a Cassie controller that joins virtual-constraint motion regulation with an angular-momentum pendulum model and model predictive control, including a physical moving-walkway demonstration. This is a particular implementation rather than a universal virtual-constraint architecture.

A desired relation must be feasible

Feedback cannot enforce a trajectory that conflicts with actuator limits, contact forces, joint ranges, or the robot's underactuation. Poor output selection can leave important internal motion weakly controlled even when the measured output error is small.

The phase variable also needs to progress reliably. If it reverses or becomes ambiguous during a disturbance, the desired joint values can become inappropriate. Stability and feasibility therefore depend on the model, chosen outputs, gait construction, and treatment of impacts.

Sources