Hybrid System

A hybrid system's state evolves through two distinct modes:

• Continuous Flow: Described by differential or difference equations (e.g., dx/dt = f(x, u)). This governs motion between events, like a robot arm swinging.
• Discrete Jumps: Triggered by events or guards, causing an instantaneous state reset (e.g., x^+ = g(x^-)). This models actions like making/breaking contact or changing control modes. The interplay defines behaviors like bouncing balls (continuous fall, discrete bounce) or thermostats (continuous cooling, discrete heater switch).

These three components formally define discrete transitions:

• Guard: A condition in the continuous state space (e.g., foot_height(x) <= 0). When true, a transition is enabled.
• Reset Map: A function applied when a transition is taken, updating the state (e.g., velocity = -0.8 * velocity for a bounce).
• Invariant: A condition that must hold for the system to remain in a given discrete mode. Violation forces a transition. Together, they create a hybrid automaton, the standard mathematical model for these systems.

Hybrid systems introduce complex temporal phenomena:

• Zeno Behavior: An infinite number of discrete transitions occurring in a finite time interval (e.g., a bouncing ball whose bounce height asymptotically approaches zero). This poses major challenges for simulation and analysis.
• Non-Determinism: Arises when multiple transitions are enabled simultaneously from a state. The system must choose one, leading to branching behavior that requires supervisory control or probabilistic modeling. These features necessitate rigorous formal methods to guarantee system safety and liveness.

A quintessential application in robotics is modeling rigid-body contact.

• Continuous Phase: Dynamics governed by equations of motion (Lagrangian or Newton-Euler).
• Discrete Transition: Triggered by a zero-distance guard condition.
• Reset Map: Applies an impact law (e.g., Newton's restitution law) to compute post-collision velocities. This hybrid model is fundamental for legged locomotion (walking robots), manipulation (grasping), and any task involving making/breaking contact with the environment.

Analyzing hybrid systems requires specialized verification tools to answer critical questions:

• Reachability Analysis: Computing the set of all states a system can reach from an initial set. Is an unsafe state reachable?
• Tool Support: Frameworks like SpaceEx, Flow*, and HybridSal use techniques from computational geometry and symbolic computation to over-approximate reachable sets.
• Applications: Proving collision avoidance for autonomous vehicles, stability of switching power converters, and safety of medical device software.

Hybrid systems naturally model hierarchical control architectures common in advanced robotics.

• High-Level Planner: Operates in a discrete mode (e.g., "walk", "stand", "recover"), encoded as a finite-state machine.
• Low-Level Controller: In each mode, a continuous controller (e.g., MPC, LQR) executes.
• Mode Switching: Triggered by guards based on sensor data (e.g., "if tilted > 15°, switch to recovery mode"). This structure is central to Boston Dynamics' robots and autonomous vehicle decision-making (e.g., lane change, emergency brake).

A Differential-Algebraic Equation (DAE) system contains both differential equations (describing continuous evolution) and algebraic equations (describing constraints). This is the primary mathematical representation for the continuous dynamics phase of a hybrid system, especially when modeling robots with kinematic loops or persistent contact constraints.

• Key Role: Models the continuous flow between discrete transitions.
• Example: The equations of motion for a robot arm where joint angles are differential states, but the closed kinematic chain of a parallel manipulator introduces algebraic constraints.

A Partially Observable Markov Decision Process (POMDP) is a mathematical framework for decision-making under uncertainty where the agent cannot directly observe the true system state. It is a critical model for the discrete decision-making layer in a hybrid system, especially when perception is noisy.

• Relation to Hybrid Systems: The high-level task planner or mode selector in a robot often operates as a POMDP, choosing which discrete control mode (e.g., 'grasp', 'move', 'recover') to activate based on imperfect sensor data.
• Components: Includes a belief state (probability distribution over true states), actions, observations, and rewards.

A Control Barrier Function (CBF) is a mathematical tool used to synthesize controllers that guarantee a system remains within a safe set. It is a primary method for enforcing the invariant conditions of a discrete mode in a hybrid system, ensuring the continuous dynamics do not violate safety before a transition is triggered.

• Enforcing Mode Invariants: For a walking robot's 'left-foot-stance' mode, a CBF can formally guarantee the center of pressure remains within the foot's support polygon.
• Formal Guarantee: Provides a filter for any nominal controller, modifying control inputs in real-time to ensure the barrier condition (and thus safety) is never violated.

A Lyapunov function is a scalar energy-like function used to prove the stability of an equilibrium point for a dynamical system. In hybrid systems analysis, Lyapunov functions (and their hybrid variants) are used to prove stability across discrete transitions, ensuring the entire system converges to a desired state despite switching.

• Hybrid Stability: A common approach is to find a common Lyapunov function that decreases during continuous flow and does not increase during any discrete jump (transition).
• Application: Proving a multi-legged robot's gait controller is stable, even as it switches between flight and stance phases.

Controllability is a fundamental property of a dynamical system determining whether it is possible to steer the system from any initial state to any desired final state using admissible control inputs. For hybrid systems, this concept extends to hybrid controllability, which considers the combined effect of continuous control inputs and discrete mode transitions to achieve a goal.

• Extended Capability: Discrete mode switches can make a system hybrid-controllable even when its continuous dynamics in any single mode are not fully controllable. For example, a car (nonholonomic) can park by switching between forward and reverse gears.
• Analysis: Determines if a robotic task (e.g., reorienting a grasped object) is feasible given the available contact modes and actuator limits.

Zeno behavior is a pathological phenomenon in hybrid systems where an infinite number of discrete transitions occur in a finite amount of time. This is a critical consideration for the well-posedness and practical implementability of a hybrid controller on physical hardware.

• Cause: Often arises from poorly designed guard conditions or reset maps, such as a bouncing ball model where the bounce height approaches zero asymptotically.
• Robotics Impact: A robot controller exhibiting Zeno behavior would attempt to switch modes infinitely fast, leading to software failure or hardware damage. Formal verification must rule out Zeno executions for reliable deployment.