This work designs a runtime shield that comes with a finite-horizon guarantee: with high probability over the training data, if the true perception uncertainty rates lie within the learned intervals, then every action admitted by the shield satisfies a stated lower bound on safety.
Abstract
Autonomous systems that rely on learned perception can make unsafe decisions when sensor readings are misclassified. We study shielding for this setting: given a proposed action, a shield blocks actions that could violate safety. We consider the common case where system dynamics are known but perception uncertainty must be estimated from finite labeled data. From these data we build confidence intervals for the probabilities of perception outcomes and use them to model the system as a finite Interval Partially Observable Markov Decision Process with discrete states and actions. We then propose an algorithm to compute a conservative set of beliefs over the underlying state that is consistent with the observations seen so far. This enables us to construct a runtime shield that comes with a finite-horizon guarantee: with high probability over the training data, if the true perception uncertainty rates lie within the learned intervals, then every action admitted by the shield satisfies a stated lower bound on safety. Experiments on four case studies show that our shielding approach (and variants derived from it) improves the safety of the system over state-of-the-art baselines.
Autonomous systems under partial observability act on beliefs, not raw sensor events. QANTIS treats the quantum processor as a calibrated belief-update service in that loop: it receives a prior and an observation model, estimates the rare-event evidence term, and returns an ordinary posterior to a classical planner. This paper asks whether that service can be reused across a sequential Tiger POMDP horizon on present IBM Heron hardware without corrupting the planner-facing posterior. We answer with a controlled hardware case study rather than an end-to-end autonomy or wall-clock speedup claim. The study compares no amplification, guarded Grover amplification, and all-step fixed-point amplification on the same trajectory, then checks whether the returned posterior would change the downstream action. All-step FPAA preserves the Tiger posterior across the reported 8-step and 12-step primary runs, and the 20-step and 32-step controls remain inside the same operating band. In every reported decision check, the hardware posterior and the exact Bayes posterior select the same immediate action. Boundary-aware BIQAE stabilizes amplitude estimation near zero and near one, while a rare-event sweep maps the logical sample-complexity envelope for one-in-a-million evidence. The result is an operating envelope for a hardware-calibrated belief-update primitive, not a standalone hardware-advantage claim.
Bayram Yuksel Eker, Suayb S. Arslan, O. Nazli et al.· 0 citations
Ensuring robot safety in unknown, dynamic environments is a fundamental requirement. It involves inferring the states of an unknown and time-varying number of moving objects from noisy, incomplete measurements. We address safe control under the induced multi-object state uncertainty with a risk-aware belief control barrier function (BCBF) framework. The uncertainty is captured by a random finite set (RFS) belief, estimated by a sequential Monte Carlo probability hypothesis density (SMC-PHD) filter that represents it with a set of particles. Building directly on these particles, we construct a nonsmooth BCBF, establish forward invariance of the safe set under continuous prediction, and derive an explicit condition under which discrete updates preserve safety. Simulation and real-world underwater experiments demonstrate the effectiveness and efficiency of the proposed approach.
Shaohang Han, Gang Chen, Yixi Cai et al.· 0 citations
This lecture note develops, from first principles and within a single decision-theoretic setting, the link between the decision objective and the knowledge of an agent and the form of uncertainty representation that is sufficient to act optimally.
Model-free reinforcement learning methods for navigation often suffer from poor generalization,
sparse rewards, and failures under partial observability. This paper introduces Probabilistic
Reconstruction and Informed Search with Memory (PRISM), a hybrid framework that decouples
environment understanding from decision-making. Rather than learning a navigation policy end to end,
a generative adversarial network (GAN) reconstructs and completes a global map of the environment
from partial observations and agent memory. Known observations are strictly preserved, whereas
unknown regions are probabilistically inferred to produce plausible world hypotheses consistent with
the sensed data. A* search uses the generated map to calculate a minimum-cost path to the target under
the planner’s traversal-cost model and replans at each step as new observations update the memory. This
separation enables environment reconstruction under limited sensing and interpretable, deterministic
planning decisions. Across five estimate-level evaluations per condition, A*+GAN achieved 99.82 ±
0.04% success and averaged 11.26 ± 0.92 steps, compared with 99.60 ± 0.22% success and 12.10 ± 0.09
steps for memory-only A*. A*+GAN had lower all-episode mean steps in all five matched evaluations.
The DQN conditions averaged 7.68% to 18.92% success and showed substantially greater between-model
variation. Overall, the findings indicate that learning the structure of the world can facilitate effective
classical planning without policy memorization, providing an alternative to model-free reinforcement
learning for partially observable navigation.
Advait Johari· American Journal of Student...· 0 citations
Validating agentic artificial intelligence systems that operate through long-horizon interaction with complex environments remains a fundamental and unresolved challenge. Contemporary validation practice relies heavily on scenario-based testing, benchmarking, and large-scale simulation, implicitly assuming that sufficient scenario coverage can approximate the system’s operational behavior. This paper challenges that assumption by framing validation as an inference problem under bounded information and analyzing its limits from an information-theoretic perspective. We formalize deployment behavior as a random variable over an expansive scenario space and characterize validation as the acquisition of finite information through observed interaction trajectories. Under minimal and realistic assumptions, we show that the entropy of the scenario space grows rapidly with interaction horizon, while the information obtainable through finite validation is strictly bounded. As a consequence, exhaustive scenario coverage and complete validation guarantees are, in general, information-theoretically impossible for agentic AI systems. We further demonstrate that empirical coverage metrics may diverge from epistemic certainty, leading to increasing apparent validation confidence without proportional reduction in residual uncertainty. To address these limits constructively, we identify conditions under which scope-limited, conditional guarantees remain attainable when validation is restricted to explicitly defined scenario classes. Theory-confirming empirical illustrations are provided to demonstrate the saturation of validation capacity and the persistence of uncertainty under increasing validation effort. The results delineate principled boundaries on certifiable claims for agentic AI systems, while also clarifying the conditions under which validation evidence remains epistemically interpretable, evaluation, and governance practices.
S. S· i-manager's Journal on Compu...· 0 citations
A robot that has to see and react on a fixed clock runs into two problems at once. Its cameras degrade in rain, mud, fog, and darkness. And the single onboard processor it runs on is shared with planning and control, so the compute left over for perception moves around from second to second. Most systems model the two separately. We present a perception router that tracks probabilistic estimates of sensor-fault state and compute- contention state, couples them with a noisy-OR term, and uses the coupled estimate to pick one of four detector configurations (YOLO11x/n at 1280 or 640 px) so that the frame finishes before its deadline. Where the two stressors co-occur, the coupled policy cuts the deadline-miss rate by 1.1 to 9.4 percentage points against a policy that treats them independently. The interval excludes zero in five of six conditions, the pooled effect over 10 sequences and 6 conditions has sign-test p = 0.001, and every uncoupled control and the fault-free trajectory sit at exactly 0.0 pp. Routing costs tens of microseconds per frame. We then asked whether the coupling the method exploits arises on its own. Across eight real RADIATE adverse-weather sequences and three workload proxies independent of the fault signal, after Benjamini-Hochberg correction and a replication run, none of 24 tests found it. We report that null and scope the routing result as a proof of mechanism. Whether such coupling occurs in the field is still open, and the released evaluation pipeline lets a deployment settle it on its own traces.
