online replanning

This work addresses the challenge of intractable exact solution for partially observable Markov decision processes (POMDPs) due to their high computational complexity by proposing an adaptive open-loop simplification framework. The approach constructs a belief tree based on topological structure and alternates between open-loop and closed-loop planning. It introduces, for the first time, a safety-aware replanning-skipping mechanism for multi-step open-loop action sequences with formal performance guarantees. By deriving efficiently computable performance bounds, the method ensures that the simplified planning process still identifies the optimal immediate action of the original problem. Experimental results demonstrate that the proposed framework significantly reduces planning overhead while preserving provable performance guarantees, thereby substantially improving the scalability and efficiency of online POMDP solvers.

This work addresses the problem of active information regulation planning for agents operating in partially observable environments while aware of an observer. To overcome the limitation of existing observer-aware models—which assume full observability—we propose the first Partially Observable Observer-Aware Markov Decision Process (PO-OAMDP), capable of handling dynamic hidden states and runtime goal changes. Methodologically, we integrate Partially Observable MDP (POMDP) modeling with Heuristic Search Value Iteration (HSVI), augmented by a customized initialization strategy to accelerate convergence. We theoretically analyze structural properties of PO-OAMDP and empirically demonstrate that our framework significantly improves policy legibility and predictability—key desiderata for explainable AI—while verifying enhanced HSVI convergence on benchmark tasks. The core contribution is the principled extension of observer-aware decision-making to partial observability, enabling controllable, interpretation-aware information disclosure.

Real-time robot replanning in dynamic environments incurs high computational overhead and relies heavily on explicit change detection and graph updates. Method: This paper proposes a novel incremental planning paradigm that eliminates the need for explicit reuse or update of historical paths. It decouples dynamic replanning into a sequence of independent, asymptotically optimal sampling-based planning problems—thereby avoiding dependence on obstacle change perception and dense graph reconstruction. The approach leverages almost-surely asymptotically optimal algorithms, including Effort-Informed Trees* (EIT*) and Asymptotically Optimal RRT-Connect (AORRTC), to balance rapid initial solution generation with continuous path refinement. Contribution/Results: Simulation results show that EIT*-generated paths achieve significantly shorter median lengths than those produced by mainstream reactive planners. Physical experiments on a robotic manipulator demonstrate AORRTC’s effectiveness and robustness in complex, dynamic task scenarios.

This paper addresses the problem of minimizing modifications to the initial state—termed “plan perturbation”—while achieving a given goal in automated planning. We formally define plan perturbation for the first time and propose a multi-objective optimization framework jointly minimizing action execution cost and state perturbation magnitude. Leveraging planning compilation techniques, we embed this bi-objective optimization into classical planning solvers, enabling integrated modeling and principled trade-offs between action costs and state changes. Experiments across multiple benchmark domains demonstrate that our approach efficiently generates feasible plans with low perturbation, bounded action cost, and semantic smoothness—significantly outperforming conventional planners optimizing action cost alone. Our core contributions are threefold: (1) a computationally grounded formal definition of plan perturbation; (2) a compilable, scalable multi-objective planning framework; and (3) empirical validation of its effectiveness and robustness in realistic scenarios.

Discrete partially observable Markov decision processes (POMDPs) lack deterministic performance guarantees for online planning solutions. Method: This paper proposes an online planning framework that, for the first time, establishes a deterministic error bound between any time-bounded approximate solution and the optimal value function. The approach integrates POMDP modeling, online Monte Carlo tree search (MCTS), upper-confidence-bound propagation, and rigorous error bound derivation—designed as a plug-and-play enhancement to existing MCTS-based planners. Contribution/Results: Evaluated on standard benchmarks, the method achieves significantly improved solution quality with negligible increase in computational overhead, while providing verifiable theoretical guarantees. It is the first POMDP planning framework to simultaneously ensure real-time execution, deterministic error bounds, and plug-and-play compatibility with state-of-the-art tree search algorithms.

This work addresses the challenge that distributional shifts in real-world environments often cause conventional planning models to fail under partial observability. The authors propose a causal POMDP framework that, for the first time, models distributional shifts as causal interventions. Within an augmented belief space, the approach jointly infers latent states and domain changes, enabling policy evaluation and adjustment under hypothetical interventions. Theoretical analysis demonstrates that the value function retains its piecewise-linear and convex structure in this extended space, ensuring the solvability and computational tractability of α-vector–based planning. By actively identifying components of environmental change, the method significantly enhances adaptability in dynamic, non-stationary scenarios.

This work addresses the challenge in long-horizon agent planning where local errors tend to propagate and amplify through the planning graph, while conventional global replanning incurs prohibitive computational costs. To mitigate this, the authors propose WM-SAR, a method that identifies root causes by backtracking error-amplifying subgraphs and selectively feeds only causally relevant subgraphs into a large language model (LLM) for in-situ repair, thereby avoiding full-graph replanning. Integrating graph-structural analysis, subgraph amplification mechanisms, and LLM-based reasoning, WM-SAR substantially reduces context consumption while enhancing repair accuracy. Experimental results demonstrate that under constrained token budgets, WM-SAR significantly outperforms symptom-scanning engineering correctors, achieving planning stability comparable to full-graph replanning through compact subgraphs and offering more precise repair targets.

This work addresses the optimal observability problem (OOP) in uncertain environments, which entails balancing task feasibility against sensing costs. Focusing on its decidable subproblems—sensor selection (SSP) and position observability (POP)—the paper proposes a novel solution framework based on POMDP decomposition, integrating parameter synthesis with a symbolic–subsymbolic hybrid approach. This method dramatically improves computational efficiency, scaling solvable instances by three orders of magnitude and reducing runtime by five orders of magnitude compared to prior techniques. Consequently, the approach substantially expands the tractable boundary of observability-aware planning in partially observable settings.

This work addresses safe navigation in dynamic environments with uncertain, time-varying obstacles by anticipating local observations. It introduces the first integration of precise contingency planning with Safe Interval Path Planning (SIPP) to generate formally verified safe macro-actions. The approach performs bounded AND/OR search over a cached action–observation graph to select optimal action sequences for each reachable observation. To guide search efficiently, it employs optimistic and robust SIPP relaxations that yield admissible heuristic bounds. Decisions are made dynamically based on local observations, enabling real-time adaptation. Experiments demonstrate superior performance over fixed-path baselines in controlled road networks and successful planning in gated scenarios where conservative methods fail. The study also reveals a scalability bottleneck as observation uncertainty increases.

This work addresses the performance degradation of policies in partially observable Markov decision processes (POMDPs) under observation model mismatches, such as sensor degradation, by formally introducing the notion of *policy observation robustness*: the largest perturbation range of the observation model under which a policy’s value is guaranteed to remain above a specified threshold. The authors develop bilevel optimization frameworks tailored to sticky and non-sticky observation biases, proving that in the non-sticky case, only finite-state controller (FSC) nodes need be considered. They propose the Robust Interval Search algorithm, which enjoys completeness and convergence guarantees, scales efficiently to POMDPs with tens of thousands of states, and achieves polynomial time complexity in the non-sticky setting. Empirical validation in robotics and operations research demonstrates its practical utility.