Dynamic Constraint Reconstruction Based Control Barrier Functions for Safety-Critical Control of High-Dimensional Manipulators
arXiv:2607.15961
2026
Dynamics
2 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper provides a transferable disturbance-reconstruction mechanism: an extended state observer estimates the lumped dynamics error online, and the estimate is inserted into a high-order control barrier function rather than treating uncertainty only through a fixed worst-case bound. A safety margin based on observer error yields a sufficient forward-invariance condition, allowing the controller to be less conservative while retaining zero-violation guarantees. The direct neural-network transfer is a differentiable or QP-based safety layer around a learned policy, with the observer state and adaptive margin making constraints depend on currently inferred disturbances.
Ideas from this paper
✗ Failed on benchmark
2026
Wrap a neural policy with a control-barrier safety layer whose constraints use an online estimate of model mismatch or environmental disturbance. Instead of enforcing a fixed worst-case bound at every state, the layer reconstructs the current effective dynamics from an extended state observer and adds only the margin required by the remaining estimation error.
Useful8/10
Difficulty6/10
Novelty6/10
✗ Mechanism failed
2026
Train a neural policy against the same dynamically reconstructed barrier used during inference. Penalize barrier violations using the current observer uncertainty margin, causing the policy to avoid states where safety would require large corrective projections.
Useful7/10
Difficulty5/10
Novelty7/10