Hybrid system
- Also called: hybrid dynamical system
- Category: Control › Feedback control
- Tags: dynamics, hybrid-dynamics, contact, locomotion, optimal-control
A dynamical system that combines continuous dynamics with discrete switches between modes, such as a legged robot alternating between stance and flight, or a hand making and breaking contact.
Definition
A hybrid system has a state that evolves continuously within each of several discrete modes and jumps between modes when certain conditions are met. Each mode has its own continuous dynamics, usually differential equations. A guard, a surface or region of the state space, triggers a transition when the state reaches it, and a reset map gives the state immediately after the transition, which may be discontinuous. A standard example is a bouncing ball: free flight is the mode, the guard is the ball reaching the floor while moving down, and the reset reverses and scales down the vertical velocity.
Robots are full of hybrid behavior. A walking robot changes dynamics every time a foot touches down or lifts off, and impacts cause near-instantaneous velocity changes. A manipulator pushing an object switches between sticking and sliding contact. A VTOL aircraft moves between hover, transition and cruise flight, each with different aerodynamics. Discrete logic in a controller, such as a state machine that switches behaviors, also makes the closed-loop system hybrid.
The discrete jumps break tools that assume smooth dynamics. The sensitivity of a trajectory to its initial conditions or controls changes abruptly across a guard, and the saltation matrix gives the correct linearization through a reset. Optimization methods handle hybrid systems in several ways: by fixing the mode sequence and optimizing each phase with its duration as a variable, by encoding mode choices with integer variables in mixed-integer programs, or, for contact, by contact-implicit formulations that express switching through complementarity constraints. Stability analysis of periodic behaviors such as gaits uses Poincaré maps, and some models exhibit Zeno behavior, an infinite number of switches in finite time, which simulators must guard against.
Why it matters
Locomotion, manipulation and many flying vehicles are hybrid. Treating them as smooth leads to controllers and optimizers that fail at the switching events, so recognizing the hybrid structure decides which planning, estimation and stability tools are valid.
Key applications
- Legged locomotion planning and control
- Contact-rich manipulation planning
- Control of VTOL and transitioning aircraft
- Stability analysis of periodic gaits
- Verification of robots driven by supervisory state machines
In practice
A team developing a one-legged hopping robot models it as a hybrid system. In the flight mode the body follows ballistic dynamics; in the stance mode it follows spring-leg dynamics. The touchdown guard fires when the foot reaches the ground, the liftoff guard when the leg spring returns to its rest length, and the touchdown reset removes the velocity lost in the foot impact. They fix the mode sequence and optimize the stance trajectory together with the phase durations. To linearize the motion for a stabilizing controller they use the saltation matrix at touchdown instead of finite differences across the event, and in simulation they use event detection to locate guard crossings precisely, since a fixed-step integrator that misses the exact contact time introduces errors.
Common misconceptions
- Hybrid means combining two control methods — in dynamics, it means continuous evolution mixed with discrete mode switches, unrelated to hybrid position/force control or hybrid electric drives
- Switching can be smoothed away without consequence — smoothing can help optimizers, but it changes the model, and impacts and stick-slip transitions still need care in the final controller
- Hybrid systems are rare outside legged robots — any robot that makes or breaks contact, or runs a mode-switching controller, is hybrid
Who should know this
- Legged robot control engineers
- Manipulation planning researchers
- Engineers working on transitioning aircraft
- Simulation developers implementing contact and event handling
Related technologies
- Hybrid Automata
- Saltation Matrix
- Poincaré Map
- Complementarity Constraints
- Event Detection in ODE Solvers
Related terms
- Contact mode
- Spring-Loaded Inverted Pendulum
- Mixed-Integer Programming
- Gait
- Differential dynamic programming
- Region of attraction
Related projects
- Open Duck Mini V2: An open-source mini duck robot you 3D-print and build, inspired by Disney's BDX droid
- Drake: MIT robotics toolbox for model-based dynamics, optimization, and control system design
- mjlab: Isaac Lab API on top of GPU-accelerated MuJoCo Warp for RL and robotics research
- ROBOTO ORIGIN: RoboParty's open-source 1.25 m bipedal humanoid with 23 DOF and a full assembly manual
- ASAP: Sim-to-real framework for agile humanoid whole-body skills using delta action models
- Berkeley Humanoid Lite: Open-source sub-$5K humanoid robot with 3D-printed gearboxes and full sim-to-real pipeline