- Author
- Johannes Autenrieb (German Aerospace Center, DLR), Peter A. Fisher and Anuradha Annaswamy (MIT)
- Published
- 18 June 2026
- Kind
- Control theory preprint, with proofs and one simulation
- Original
- arxiv.org/abs/2603.14403
- Subject
- Mathematics and optimization
What it is
Autenrieb, Fisher and Annaswamy extend earlier work on reference-based safe adaptive control. They treat plants where both the dynamics and the actuator effectiveness are uncertain. The aim is to keep the plant in a safe set during adaptation.
What it says
- The paper says adaptive control gives stable tracking but not safety during adaptation.
- The authors place a barrier-function safety filter on the reference signal, not on the plant.
- They use Lipschitz bounds to derive a condition that depends on tracking error.
- The condition becomes a convex second-order cone program for the safety filter.
- The paper says the constraints loosen as the plant-reference tracking error shrinks.
- In a quadrotor obstacle test, plain plant-level and reference-level filters collided with the obstacle.
- The two adaptive-CBF filters stayed safe but were more oscillatory than the proposed filter.
What evidence it gives
The paper gives a stability theorem and a safety theorem, each with a proof. It then reports one MATLAB simulation of a quadrotor avoiding a spherical obstacle. Five filters are compared by plotted trajectories, with no numeric table. The authors report that their filter stayed safe and gave smoother paths.
Interests and limits
The authors work on adaptive control and acknowledge Boeing support for one author. The paper's claim of smoother paths and less conservatism rests on plotted trajectories alone. The test is a single simulation with static model errors. No hardware flight is reported. The theory needs assumed known bounds on parameter errors and Lipschitz constants.