Predictor-Compensated CLF-QP for Vehicle Tracking with Sensing and Actuation Delays
Delay-aware control for teleoperated vehicle tracking using CLF-QP with SQP.
Overview
This project studies delay-aware vehicle tracking for teleoperation. The controller combines predictor compensation with a control Lyapunov function quadratic program (CLF-QP) solved via SQP to enforce stability and constraints under sensing and actuation delays.
Agenda
- Background and motivation
- Problem statement
- Literature review
- Algorithm overview
- Results and discussion
- Summary and future work
Background and Motivation
Modern control systems experience delays from sensing pipelines, communication networks, computation, and actuator dynamics. In safety-critical settings like autonomous driving, teleoperation, V2X communication, and multi-vehicle coordination, these delays can degrade tracking and compromise stability.
Existing methods have gaps when measurement delays and actuation delays occur simultaneously.
- Predictor feedback ensures stability but does not directly enforce stability constraints.
- Delay-aware barrier approaches emphasize invariance but lack explicit performance guarantees and often treat only a single delay channel.
Problem Statement
Objective: Compute an optimal actuation command using CLF-SQP under state and input delays.
System and Optimization
- Delay-aware system model with sensing and actuation delays
- Optimization problem with box and rate constraints
Optimization Solver: SQP
Decision Variables
- Control inputs and slack variables
Constraints and Bounds
- Amplitude and rate limits
- Rate limits referenced to previous command
- CLF inequality constraint
- No equality constraints
Objective
- Quadratic cost with tracking and effort penalties
Warm Start
- Initialize with the previous solution to reduce iterations
Algorithm
- Predictor estimates the arrival state
- CLF-QP computes control consistent with Lyapunov decrease
- SQP solves a small QP each cycle
Simulation Setup
- Teleoperation tracking scenario with known and unknown delays
Results and Discussion
True Delay Known
- Stable tracking with CLF-QP
- Constraint compliance under delay
True Delay Unknown
- CLF-QP remains resilient to delay estimation bias
- Predictive control preserves stability trends
Conclusion
- Objective and guarantees: CLF-QP enforces Lyapunov decrease directly, providing an explicit stability certificate under delay.
- Constraint handling: CLF-QP respects amplitude and rate limits through box constraints; PID saturates after the fact.
- Delay handling: Both approaches use a predictor, but CLF-QP solves for commands consistent with the predicted arrival state and Lyapunov decrease.
- Computation: The SQP-based QP is small and converges reliably at high update rates (e.g., 200 Hz).
Summary of Learning
- Lyapunov-centric objective yields formal stability decay; slack controls strictness.
- With PSD/PD weights and affine constraints, the QP remains convex and small.
- Slack preserves feasibility and tunes the trade-off between stability and constraint satisfaction.
Future Work
- Joint CLF-CBF optimization to enforce performance and safety simultaneously
- Robustness to model uncertainty and disturbances
- Extend from kinematic bicycle to nonlinear dynamics with adapted predictor and constraints
Acknowledgments
Thank you. Questions welcome.