# LLM-guided Planar Loco-Manipulation for Quadruped Robots

DevFeed: [LLM-guided Planar Loco-Manipulation for Quadruped Robots](<https://devfeed.tech/articles/llm-guided-planar-loco-manipulation-for-quadruped-robots-108883.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.40248>)

Author: Georges Jetti, Michael Khayyat, Jean-Pierre Sleiman, Stefano Arrigoni, Francesco Braghin

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Robot Manipulation](<https://devfeed.tech/topics/robot-manipulation.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Robotics Simulation](<https://devfeed.tech/topics/robotics-simulation.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [llm](<https://devfeed.tech/tags/llm.md>), [lora](<https://devfeed.tech/tags/lora.md>), [manipulation](<https://devfeed.tech/tags/manipulation.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [planner](<https://devfeed.tech/tags/planner.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [robots](<https://devfeed.tech/tags/robots.md>), [sequences](<https://devfeed.tech/tags/sequences.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [structured](<https://devfeed.tech/tags/structured.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>)

## AI overview

The system plans interactive navigation by jointly choosing a skill, an object, and a geometric target, then executes variable-length parameterized action sequences with low-level reinforcement-learning policies. Its language-conditioned planner is LoRA-fine-tuned on synthetic data and outperforms a classical planner and a prompted frontier model on interactive navigation tasks. With structured scene representations, models trained from scratch can compose actions without pretraining, while pretraining gives more reliable geometric targets. An end-to-end whole-body planar pushing policy reaches 93.3% simulation success versus 83.5% for a hierarchical baseline and cuts completion time by over 60%. On hardware, the hierarchy selects the correct skill composition in 44 of 45 trials, with end-to-end success in 34 of 45.

## Source excerpt

arXiv:2610.40248v1 Announce Type: cross Abstract: Robotic navigation typically assumes a collision-free route to the goal exists; in cluttered environments with movable obstacles, however, reaching the goal may require physical interaction. Classical solutions encode these behaviors through geometric rules and task-specific logic, while learning-based approaches often constrain the decision space to low-dimensional motion commands or predefined action sets. We instead consider a broader decision space, jointly selecting the skill, relevant object, and geometric target. Our language-conditioned task planner, fine-tuned with LoRA on synthetic data, maps natural-language tasks and a structured scene representation to variable-length sequences of parameterized actions, executed by low-level reinforcement-learning policies. The planner learns the geometric relations underlying interactive navigation and outperforms a classical planner and a prompted frontier model on interactive navigation tasks. Comparisons with models trained from scratch indicate that composition does not require pretraining when the representation is structured, whereas pretraining yields more reliable geometric targets. At the low level, we introduce an end-to-end whole-body planar pushing policy that achieves a 93.3% success rate in simulation, versus 83.5% for a hierarchical baseline, while reducing completion time by more than 60%. On hardware, the hierarchy yields the correct skill composition in 44/45 trials and end-to-end success in 34/45. Project website: https://anonymousprojects-dev.github.io/LLM-guided-LocoManipulation-Quadrupeds/