Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Cognitive Robotics with Commonsense through LLM-Generated Preconditions and Subgoals

Ohad Bachner, Bar Gamliel

TL;DR

The paper tackles the failure of classical symbolic planning in robotics due to implicit commonsense knowledge by introducing a solver-first hybrid planner that injects LLM-generated preconditions and subgoals into a Unified Planning Framework (UPF). It presents a three-branch planning workflow with cache-based learning and three specialized LLM modes (REVIEW_COMMONSENSE, GENERATE_FIXED_PLAN, GAP_ANALYSIS_FOR_DOMAIN) to detect omissions, repair domains, or diagnose unsolvability. Through microwave and beer-from-fridge experiments, it demonstrates how LLM-derived deltas improve plan validity, robustness, and alignment with everyday norms, and it extends this with an end-to-end Vision-to-Execution loop in Isaac Sim to show zero-shot perception, planning, and execution. A hard benchmark and ablations illustrate that, when augmented with LLM commonsense, classical planning remains fast and scalable while producing more reliable, normative plans in dynamic environments. These results highlight the practical potential of integrating LLM commonsense with symbolic planning for resilient autonomous robotics.

Abstract

Robots often fail at everyday tasks because instructions skip commonsense details like hidden preconditions and small subgoals. Traditional symbolic planners need these details to be written explicitly, which is time consuming and often incomplete. In this project we combine a Large Language Model with symbolic planning. Given a natural language task, the LLM suggests plausible preconditions and subgoals. We translate these suggestions into a formal planning model and execute the resulting plan in simulation. Compared to a baseline planner without the LLM step, our system produces more valid plans, achieves a higher task success rate, and adapts better when the environment changes. These results suggest that adding LLM commonsense to classical planning can make robot behavior in realistic scenarios more reliable.

Enhancing Cognitive Robotics with Commonsense through LLM-Generated Preconditions and Subgoals

TL;DR

The paper tackles the failure of classical symbolic planning in robotics due to implicit commonsense knowledge by introducing a solver-first hybrid planner that injects LLM-generated preconditions and subgoals into a Unified Planning Framework (UPF). It presents a three-branch planning workflow with cache-based learning and three specialized LLM modes (REVIEW_COMMONSENSE, GENERATE_FIXED_PLAN, GAP_ANALYSIS_FOR_DOMAIN) to detect omissions, repair domains, or diagnose unsolvability. Through microwave and beer-from-fridge experiments, it demonstrates how LLM-derived deltas improve plan validity, robustness, and alignment with everyday norms, and it extends this with an end-to-end Vision-to-Execution loop in Isaac Sim to show zero-shot perception, planning, and execution. A hard benchmark and ablations illustrate that, when augmented with LLM commonsense, classical planning remains fast and scalable while producing more reliable, normative plans in dynamic environments. These results highlight the practical potential of integrating LLM commonsense with symbolic planning for resilient autonomous robotics.

Abstract

Robots often fail at everyday tasks because instructions skip commonsense details like hidden preconditions and small subgoals. Traditional symbolic planners need these details to be written explicitly, which is time consuming and often incomplete. In this project we combine a Large Language Model with symbolic planning. Given a natural language task, the LLM suggests plausible preconditions and subgoals. We translate these suggestions into a formal planning model and execute the resulting plan in simulation. Compared to a baseline planner without the LLM step, our system produces more valid plans, achieves a higher task success rate, and adapts better when the environment changes. These results suggest that adding LLM commonsense to classical planning can make robot behavior in realistic scenarios more reliable.

Paper Structure

This paper contains 39 sections, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Classical Planning and Heuristic Search
  4. LLMs and Symbolic Models
  5. Generalization and World Models
  6. Architecture: A Solver-First Hybrid Planner
  7. Core Planning Logic
  8. Branch 1 & 2: Cache-First Execution
  9. Branch 3A: Solver Success (The "Beer Case")
  10. Branch 3B: Solver Failure (The "Microwave Case")
  11. LLM Interaction Modes
  12. Runtime and Complexity Analysis
  13. Design Choices
  14. Deploying LLM commonsense in different variations
  15. Microwave Task: LLM Detects Missing Action
  16. ...and 24 more sections

Figures (3)

  • Figure 1: Planner flow with LLM-induced subgoals feeding back into the plan.
  • Figure 2: This is the image of our Robot Franka that our Vision LLM got.
  • Figure 3: system flow