- The paper presents a set-theoretic fast ellipsoidal MPC that replaces intensive online non-convex optimization with efficient set-membership evaluation.
- It guarantees recursive feasibility and strict safety margins through precomputed robust invariant ellipsoidal tubes despite dynamic obstacles and uncertainties.
- Experimental validation in MATLAB/Simulink and CARLA shows significant improvements in computational efficiency and trajectory accuracy over traditional NLMPC methods.
Set-Theoretic Receding Horizon Control for Robust Overtaking in Autonomous Highway Driving
Introduction and Problem Motivation
The paper "Set-Theoretic Receding Horizon Control for Obstacle Avoidance and Overtaking in Autonomous Highway Driving" (2604.01790) introduces a formal control-theoretic approach to achieve real-time, robust, and constraint-satisfying motion planning for autonomous vehicles in high-speed highway environments. Overtaking maneuvers are operationally critical, requiring simultaneous regulation of lateral and longitudinal dynamics under uncertainty due to dynamic obstacles and significant modeling inaccuracies. The prevalent Model Predictive Control (MPC) methodologies, notably Nonlinear MPC (NLMPC) and its stochastic or mixed-integer variants, are limited by non-convex online optimization and scalability issues, thus impeding their use in safety-critical systems with fast sampling rates and strict actuator and state constraints.
This work proposes a set-theoretic fast ellipsoidal MPC (ST-FE-MPC), which transitions from heavy online non-convex optimization to efficient set-membership evaluation exploiting precomputed ellipsoidal inner approximations of controllable sets. The approach guarantees strict constraint satisfaction and recursive feasibility in real time, handling vehicle models with polytopic uncertainty.
Methodological Contributions
The authors design a hierarchical RHC framework embedding a set-theoretic procedure that constructs, offline, a sequence of nested ellipsoidal robust invariant sets, each representing the one-step robustly controllable region towards a local equilibrium along a planned path. The path itself is described as a sequence of way-points, with feasible connecting segments computed via backward propagation using robust control invariant ellipsoidal sets. The following theoretical and algorithmic innovations are critical:
- Polytopic System Approximation: The nominal nonlinear vehicle dynamics (6-DOF) are embedded in a polytopic, quasi-LPV state-space representation, parametrized by bounded scheduling variables, capturing both lateral and longitudinal degrees of freedom. This models both the core vehicle dynamics and structured uncertainty from traffic disturbances and model-order reduction.
- Ellipsoidal Tube Construction: Using maximum-volume, inner ellipsoidal approximations, the feasible state space is partitioned into robust invariant tubes, each guaranteeing that the vehicle can be steered to the next tube in one step, under all admissible system uncertainties and disturbances.
- Set-Membership Online Computation: At each control step, the vehicle state is mapped to the smallest tube containing it using a lexicographical search, and the regulation to the next tube is cast as a constrained min-max control problem over the polytopic vertices, cost-weighted by the inverse tube matrix, ensuring minimum-time-like contraction along the path.
- Dynamic Replanning: The approach natively admits fast replanning in response to newly detected obstacles, as tubes can be updated by reconstructing the ellipsoidal sequences to new way-points, without requiring full online optimization or violating previous feasibility guarantees.
Theoretical Properties and Guarantees
The use of ellipsoidal robust control invariant sets leads to a formal guarantee of recursive feasibility, i.e., any initial condition inside the outermost ellipsoid is steered toward the goal without violating input or state constraints, despite bounded disturbances. The framework further enforces collision avoidance rigorously, as safety margins in the tube construction account for perception noise and physical vehicle envelopes in both relative position and velocity. The proposed controller is distinctly non-conservative due to the tightness of inner ellipsoidal approximations, bridging the gap between mathematical robustness and practical disturbance rejection.
Experimental Validation and Evaluation
The paper assesses the performance in two stages: (1) simulation in MATLAB/Simulink with the Automated Driving Toolbox (ADT) and (2) high-fidelity co-simulation in the CARLA 3D simulator.
Two scenarios are considered:
- Standard Highway Overtaking: The ego vehicle performs a left-lane overtaking maneuver of a slower lead vehicle, starting with a safe following mode and switching to overtaking upon feasibility detection.
- Multi-Vehicle Dynamic Overtaking: The ego vehicle adapts its trajectory in real time in response to an additional dynamic obstacle, necessitating a lane sequence change and real-time trajectory replanning.
Key performance indicators include:
- Input Feasibility: Steering and acceleration commands are strictly confined within physical bounds, eliminating chattering and controller-induced actuator saturation.
- Safety Under Disturbances: The system enforces a non-violation of the minimum relative distance constraint, functioning as a certified TTC barrier under all scenarios tested.
- Real-Time Computational Efficiency: Online computational time is reduced by over 90% compared to NLMPC for both scenarios (mean online time per step is <9% of the 100ms sampling window), validating deterministic real-time feasibility.
- Trajectory Accuracy: Root Mean Square Errors (RMSE) for lateral/yaw tracking and position converge to zero in both static and dynamic disturbance environments; the NLMPC baseline is significantly outperformed under multi-obstacle and high-disturbance settings.
The results demonstrate that the ST-FE-MPC control structure can smoothly adapt to major dynamic disturbances and real-world modeling errors, with no violation of safety-critical physical or comfort constraints.
Implications and Broader Impact
The proposed method represents a significant step toward real-time, mathematically certified motion planning for autonomous vehicles with explicit safety and input guarantees, addressing the persistent limitation of most optimization-based MPC schemes. The use of set-theoretic offline computations (ellipsoidal sequence construction) is particularly attractive for embedded automotive hardware implementations, where deterministic real-time operation is mandatory, and for integration with formal verification flows in automotive software stacks.
From a theoretical standpoint, this framework robustifies motion planning against exogenous uncertainty using only mild approximations (ellipsoidal inner bounds), thus bridging the gap between verification theory and practical controller synthesis. The method is not restricted to overtaking but extensible to a broad class of obstacle avoidance and multi-agent dynamic planning problems in structured environments.
Future Directions
Potential future work, as indicated by the authors, involves extending the set-theoretic receding horizon design to unstructured urban scenarios, incorporating adversarial behavior predictions, and introducing explicit Fault-Tolerant Control (FTC) layers to handle severe actuator degradations. The methodology is naturally compatible with integration into hierarchical planning stacks that combine learning-based global policies with formal, set-theoretic local tracking, thus facilitating research into hybrid model-based/data-driven autonomous driving architectures.
Conclusion
This paper formalizes and demonstrates a set-theoretic, ellipsoidal tube-based receding horizon control strategy for autonomous vehicle motion planning under dynamic, uncertain, and constrained environments. The framework exhibits an overview of theoretical robustness, computational efficiency, and practical safety compliance, validated through challenging high-fidelity simulations aligned with real-world traffic scenarios. Its practical adoption could improve both the predictability and verifiability of autonomous highway driving controllers in safety-critical deployments.
Reference:
"Set-Theoretic Receding Horizon Control for Obstacle Avoidance and Overtaking in Autonomous Highway Driving" (2604.01790)
