Slide 1 – Title Bursting in Adaptive AI Controllers: A Damped-Oscillation Failure in Robotics Name, Course, Date Slide 2 – Objective & Background Link damped/forced oscillation theory to AI-based adaptive control in robots. Present bursting: a stable controller suddenly causes large oscillations. Oscillator model (conceptual): 𝑚 𝑥 ¨ + 𝑐 𝑥 ˙ + 𝑘 𝑥 = 𝐹 ( 𝑡 ) m x ¨ +c x ˙ +kx=F(t) Adaptive/AI controller changes effective 𝑐 c (damping) and 𝑘 k (stiffness) online. Slide 3 – The System: AI Controller + Physical Plant Plant: robotic arm, legged robot, or autonomous vehicle (mechanical oscillator). Controller: Learns or adapts dynamics (mass, damping, friction) using ML or adaptive laws. Commands forces/torques to track trajectories or stabilize motion. Goal: good tracking and stability under changing conditions. Slide 4 – Failure Mode: Bursting Into Oscillation Observed behavior: System runs stably for a long time. Suddenly, positions/velocities show large-amplitude oscillations (“bursting”). May later appear stable again. Consequences: Actuator saturation, mechanical stress, possible hardware damage. Loss of tracking, unsafe robot behavior. Slide 5 – Why This Is an Oscillation/Damping Failure Closed-loop dynamics behave like an oscillator with time-varying damping 𝑐 ( 𝑡 ) c(t). Parameter drift / poor estimation → effective damping 𝑐 eff c eff becomes too small or negative. System transitions from: Overdamped/stable → underdamped or negatively damped → growing oscillations. Analogy: shock absorbers that gradually lose viscosity, then effectively “push” instead of resisting motion. Slide 6 – Lessons & Mitigation (AI/Control Design) AI controllers must enforce minimum damping and stability margins, not just minimize error. Mitigation strategies: Constrain learned/estimated damping to stay above a safe threshold. Regularize dynamics models to avoid negative-damping predictions. Use stability-constrained RL / control (penalize unstable closed-loop poles). Ensure sufficient excitation/data so parameters don’t drift. Takeaway: In AI for physical systems, oscillation theory is a safety tool, not just theory.
Slide 1 – Title Bursting in Adaptive AI Controllers: A Damped-Oscillation Failure in Robotics Name, Course, Date Slide 2 – Objective & Background Link damped/forced oscillation theory to AI-based adaptive control in robots. Present bursting: a stable controller suddenly causes large oscillations. Oscillator model (conceptual): 𝑚 𝑥 ¨ + 𝑐 𝑥 ˙ + 𝑘 𝑥 = 𝐹 ( 𝑡 ) m x ¨ +c x ˙ +kx=F(t) Adaptive/AI controller changes effective 𝑐 c (damping) and 𝑘 k (stiffness) online. Slide 3 – The System: AI Controller + Physical Plant Plant: robotic arm, legged robot, or autonomous vehicle (mechanical oscillator). Controller: Learns or adapts dynamics (mass, damping, friction) using ML or adaptive laws. Commands forces/torques to track trajectories or stabilize motion. Goal: good tracking and stability under changing conditions. Slide 4 – Failure Mode: Bursting Into Oscillation Observed behavior: System runs stably for a long time. Suddenly, positions/velocities show large-amplitude oscillations (“bursting”). May later appear stable again. Consequences: Actuator saturation, mechanical stress, possible hardware damage. Loss of tracking, unsafe robot behavior. Slide 5 – Why This Is an Oscillation/Damping Failure Closed-loop dynamics behave like an oscillator with time-varying damping 𝑐 ( 𝑡 ) c(t). Parameter drift / poor estimation → effective damping 𝑐 eff c eff becomes too small or negative. System transitions from: Overdamped/stable → underdamped or negatively damped → growing oscillations. Analogy: shock absorbers that gradually lose viscosity, then effectively “push” instead of resisting motion. Slide 6 – Lessons & Mitigation (AI/Control Design) AI controllers must enforce minimum damping and stability margins, not just minimize error. Mitigation strategies: Constrain learned/estimated damping to stay above a safe threshold. Regularize dynamics models to avoid negative-damping predictions. Use stability-constrained RL / control (penalize unstable closed-loop poles). Ensure sufficient excitation/data so parameters don’t drift. Takeaway: In AI for physical systems, oscillation theory is a safety tool, not just theory.
Created using ChatSlide
This project explores frame bursting as a control problem by defining damped and forced oscillation theory, modeling robots with mass-damping-stiffness dynamics, and connecting AI adaptation to failure. It traces the AI-robot failure mechanism through a plant, adaptive controller, and feedback loop, illustrating the transition from stable operation to bursting and explaining damping drift. Finally, it converts oscillation theory into safety by constraining damping, employing stability-aware...