dynamic replanning

This work addresses the inefficiency, deviation, and premature termination commonly observed in programming agents when tackling complex GitHub issues characterized by ambiguous descriptions or multi-step coordination requirements. The authors propose LivePlan, a novel framework that decouples anomaly detection from corrective guidance by employing a deterministic rule engine to monitor execution trajectories in real time and selectively invoking a large language model to generate high-level remediation instructions only when necessary. This lightweight online intervention mechanism is integrated into the SWE-agent architecture and supports multi-model collaboration. Experimental results demonstrate that LivePlan improves average resolution rates by 9.9% (up to 15.2%) on SWE-bench Verified and Pro benchmarks, at an additional cost of merely \$0.08 per instance, effectively solving challenging unsolved problems without compromising existing success rates.

This paper addresses the modeling and analysis of valid execution traces in process systems governed by precedence and response constraints. We formalize the constraint set as a partially ordered set (poset) and establish, for the first time, a bijective correspondence between precedence/response constraint systems and linear extensions of their associated posets—thereby enabling a complete combinatorial characterization of feasible traces. Building on this foundation, we develop an exact classification framework for trace sets, supporting quantitative evaluation of process utility. Our approach integrates order theory, constraint satisfaction modeling, and linear extension theory, substantially enhancing the computability and cross-system comparability of constraint-driven processes. The theoretical framework advances process mining and conformance checking by providing rigorous foundations for trace enumeration, constraint verification, and utility-aware process analysis.

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.

Existing task planning approaches suffer from limited efficiency and interpretability in detecting goal unreachability, explaining infeasibility causes, and adapting to dynamic changes in goals or constraints. This work proposes a planning framework based on relaxed Petri net reachability, integrating incremental constraint solving with invariant synthesis to enable efficient detection of infeasible plans and provide interpretable feedback, while supporting dynamic updates in sequential task planning. Experimental results demonstrate that the proposed method identifies up to twice as many infeasible scenarios as baseline approaches, generates a comparable number of invariants, matches baseline performance in single-shot planning, and significantly outperforms existing methods in sequential planning scenarios involving dynamic updates.

This work addresses the limited goal-directed execution capability of large language models in long-horizon tasks by introducing a Goal-Directed Execution (GDE) behavioral framework. The authors conduct post-training on the Qwen3.5-122B-A10B model using 363 long-horizon, multi-tool agent tasks from office scenarios, without relying on software engineering data. This approach yields a notable improvement on SWE-Bench Pro, increasing pass@1 by 5.8 percentage points. Experimental results demonstrate significant enhancements across four core GDE capabilities: goal selection, state construction, goal consistency maintenance, and environment validation. Furthermore, the model exhibits effective cross-domain transfer between office and software engineering tasks, confirming that long-horizon post-training can successfully drive the transfer of behavioral mechanisms.

This work addresses the limitation of current large language model (LLM) planning agent evaluations, which predominantly focus on task success while neglecting the dynamic impact of other agents’ responses and physical constraints in cyber-physical systems. The authors introduce the first physically verifiable benchmark platform for demand response in smart grids, evaluating the strategic efficacy of four planning architectures—predefined, sequential, hierarchical, and search-based—within a simulated environment comprising 40 heterogeneous prosumers and a radial feeder. They propose an evaluation protocol based on paired forced counterfactuals, common random responses, and event-level deadline feasibility, combined with typed policy declarations and short instruction constraints to explicitly model schedule generation, prosumer dynamics, and power flow computation. Experiments show that three architectures yield feasible, near-optimal solutions; incorporating deadline feasibility prediction reduces average regret from 90.7 to 29.0, outperforming fixed sequential strategies by 61.1%, underscoring the substantial influence of planning architecture and highlighting solution quality selection among feasible outcomes as a key challenge.

In high-density industrial environments, heterogeneous multi-robot systems are prone to path conflicts, increased waiting times, and congestion propagation due to communication delays and execution uncertainties. This work proposes the SCALE framework, which innovatively integrates robot motion characteristics into conflict resolution and constructs a generalized conjugate action-priority hypergraph (CAPH) to dynamically adjust robot priorities, enabling online generation of feasible paths and adaptive coordination. Leveraging a reactive architecture combined with an adaptive scheduling algorithm, the approach significantly reduces congestion propagation and waiting times in both simulations and a three-day real-world warehouse deployment, thereby enhancing coordination efficiency and execution robustness of heterogeneous robot fleets in complex operational scenarios.

This work addresses the challenge of efficiently leveraging symbolic patterns to guide search in numeric planning. It proposes a dynamic guidance approach based on Symbolic Pattern Planning (SPP), which incrementally generates intermediate states and refines action schemas during search. Integrated within a "planning as satisfiability" framework, the method encodes symbolic patterns and constructs state reachability formulas to enable a flexible yet sound search strategy. Theoretical analysis establishes the completeness of the approach under certain conditions, while empirical evaluation demonstrates its ability to significantly enhance solving efficiency across multiple planning strategies, effectively balancing correctness and performance.