Anticipate, Adapt, Act: A Hybrid Framework for Task Planning

Large Language Models (LLMs) can reason over complex instructions but often fail to satisfy the physical and spatial constraints required for robotic task planning. Recent LLM-based planners directly translate text into action sequences, yet they lack structured reasoning about feasibility, reachability, and logical order, resulting in invalid or incomplete plans. We present a heterogeneous multi-LLM framework that decomposes instructions into atomic reasoning tasks and allocates them to role-specialized expert agents under a token budget for real-world computational and communicational constraints. By combining role-oriented reasoning from heterogeneous agents followed by constraint-driven plan synthesis, HEART validates capability, reachability, and constraint conditions before planning and helps produce physically executable plans while maintaining efficiency. Experiments across different household benchmarks show that HEART consistently improves plan success compared to single-LLM and rule-based planners, demonstrating that heterogeneous LLM collaboration enables robust and scalable robotic task planning under resource constraints.

While robot action planning based on large language models (LLMs) has advanced remarkably, continuing a task after a physical action failure during execution remains a key challenge. For example, after failing to grasp an object, the robot may not notice that it has fallen off the table and keep trying to detect and pick it up, stalling the task. Existing approaches either rely on closed-loop autonomous re-planning, which fails when errors exceed the robot's capabilities, or on human intervention without updating the LLM's plan, hindering subsequent planning. We present a system for LLM-based robot action planning that integrates remote human intervention with feedback-based replanning. When the LLM detects an action failure, the robot requests remote assistance; an operator resolves the failure through teleoperation and reports the outcome in natural language, which is fed back to the LLM to update its plan so that the task can continue. We evaluated the proposed system on a real-world trash-collection task, comparing it with a fully autonomous method and a teleoperation-only baseline. The proposed system improves task progress over the autonomous method in this case study. For easy-to-retrieve trash, the gap from the teleoperation reference was small, while a larger gap remained for harder-to-retrieve trash. These results suggest that, in this case study, combining remote human intervention with feedback-based replanning was associated with continued task execution even in the face of otherwise unrecoverable failures. The project website is https://emergentsystemlabstudent.github.io/REPAIR/.

Advances in advanced artificial intelligence tools have sparked research in robot autonomy, but the development of such systems has largely focused on execution rather than verifying the feasibility actions planning models propose. Like general-purpose LLMs, robotics planning models carry risks: biased toward user-specified goals, they may suggest actions misaligned with scientific ethics, they may be unsafe due to an inability to "remember" prior safety risks, or they may be vulnerable to adversarial attacks on the autonomy ecosystem. We propose a LLM-driven verification layer between planning and execution to evaluate action permissibility. Our LLM-as-a-Judge ensemble combines chain-of-thought reasoning across models and synthesizes those expert judge outputs, mirroring a combination of a mixture of experts and self-consistency approach. This layer serves as middleware, gating plans from the server's planning module before they reach the MCP server and therefore the robot's low-level controls: plans are approved, rejected for reformulation, or escalated for human review. With this system, we achieve near 85% precision across accept/escalate/reject categories 97% containment of adversarial attacks, with negligible errors between accepting and rejecting tasks, and errors mostly manifesting at the escalate boundary.