End-to-end policies, from task-specific controllers to robot foundation models, learn complex behavior from raw observations but are not guaranteed to remain safe, so they are often wrapped in a safety filter. The resulting system operates under two domains: the filter's operational design domain (ODD), where its assumptions hold and safety is certified, and the policy's competence domain, where it is reliable. Filters typically fix their conservativeness regardless of the policy's competence, trading performance inside the ODD against margin beyond it. We ask whether the policy's confidence should set that level, and propose a filter-agnostic pipeline that modulates it at runtime: inside the ODD, where every setting is safe, confidence buys performance; beyond it, low confidence adds margin, which helps when it flags where the filter's model fails. Preliminary F1TENTH simulations, where this holds, show the modulated filter combining the strengths of permissive and conservative filters.
Figures & tables
Figure 1: Confidence-modulated safety filtering. A confidence monitor βtask estimates how much the task policy πtask can be trusted on the current information and modulates the conservativeness of the safety filter (fallback π\scalebox0.6\faIcon[*]shield-alt , monitor Δ\scalebox0.6\faIcon[*]shield-alt , intervention ϕ ). Adapted from Hsu et al. [11] .
Inside ODD ( μ=1.05 )
Outside ODD ( μ=0.6 )
Complete
Timeout
Crash
Lap time (s)
Complete
Timeout
Crash
Lap time (s)
No filter
15
0
45
23.6 ± 0.1
0
0
60
–
Fixed γ=0.08
60
0
0
31.4 ± 0.2
60
0
0
31.3 ± 0.2
Fixed γ=0.55
60
0
0
30.4 ± 0.2
48
2
10
31.9 ± 0.3
Modulated γ
60
0
0
30.2 ± 0.2
60
0
0
30.8 ± 0.2
Table 1: Outcomes for 60 laps per method, inside and outside of the ODD.
Figure 2: Modulated γ outside the ODD. (a) A full lap with the modulated filter, colored by γ : caution (low γ ) is applied at the sharp corners, where the policy’s confidence is low. (b) Zooms into a sharp corner. (c–d) Same corner for fixed γ=0.55 (crashes) and fixed γ=0.08 (completes).
Safety filters have been shown to be effective tools to ensure the safety of control systems with unsafe nominal policies. To address scalability challenges in traditional synthesis methods, learning-based approaches have been proposed for designing safety filters for systems with high-dimensional state and control spaces. However, the inevitable errors in the decisions of these models raise concerns about their reliability and the safety guarantees they offer. This paper presents Adaptive Conformal Filtering (ACoFi), a method that combines learned Hamilton-Jacobi reachability-based safety filters with adaptive conformal inference. Under ACoFi, the filter dynamically adjusts its switching criteria based on the observed errors in its predictions of the safety of actions. The range of possible safety values of the nominal policy's output is used to quantify uncertainty in safety assessment. The filter switches from the nominal policy to the learned safe one when that range suggests it might be unsafe. We show that ACoFi guarantees that the rate of incorrectly quantifying uncertainty in the predicted safety of the nominal policy is asymptotically upper bounded by a user-defined parameter. This gives a soft safety guarantee rather than a hard safety guarantee. We evaluate ACoFi in a Dubins car simulation and a Safety Gymnasium environment, empirically demonstrating that it significantly outperforms the baseline method that uses a fixed switching threshold by achieving higher learned safety values and fewer safety violations, especially in out-of-distribution scenarios.
Sacha Huriot, Ihab Tabbara, Hussein Sibai
Computer Science & Engineering Washington University in St. Louis
Safe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates. Safety filters separate them at action execution, but classical designs require an analytic safety function and dynamics model, and standard minimal-intervention filters are myopic to long-horizon task return because they minimize only instantaneous action deviation. Hard projections are also undefined when no safe action exists. We present FAITH, a feasibility-aware, model-free framework that approximates the optimal state-action safety value and amortizes minimal-intervention filtering with a feedforward network. The task policy optimizes the task return through the filtered dynamics, which recovers the feasible constrained problem without a competing safety term in the task-policy update. When no action satisfies the learned safety condition, the same filter approaches the action with minimum predicted peak harm. On a double integrator example and a Safety Gym environment, FAITH achieves the highest return among methods with no feasible-start violations and matches the lowest harm from infeasible starts. On a 29-DoF humanoid, it reaches a 99.95% safety rate while retaining 97% of the unfiltered return in Walking-Avoid, and obtains the highest measured safety rate in Push-Avoid by learning to sacrifice balancing and fall away from the protected region. The same policies are also demonstrated on a real-world Unitree G1 humanoid.
Safe control of humanoid robots remains challenging due to their high-dimensional dynamics, contact-rich interactions, and sensitivity to disturbances. Although reinforcement learning has enabled effective locomotion and motion tracking, learned policies can still generate unsafe actions that lead to instability or falls. In this work, we propose residual reinforcement learning as an implicit safety-filtering mechanism for safe humanoid control. Instead of relying on a single nominal policy to simultaneously balance performance, safety, and robustness, we decouple performance and safety. The nominal policy focuses solely on task performance, while a residual policy learns safety corrections. This decoupling leads to a better performance--safety Pareto trade-off and avoids the need for careful tuning of multiple competing reward terms within a single policy training. We show that the residual policy can act as an implicit safety filter.
Gechen Qu, Tong Zhang, Bike Zhang +4
University of California, Berkeley · University of California, Los Angeles