RoboGolf: Mastering Real-World Minigolf with a Reflective Multi-Modality Vision-Language Model

TL;DR

RoboGolf is a novel framework that integrates vision-language models with multi-modal perception and reflective reasoning to master real-world minigolf tasks, demonstrating advanced embodied intelligence and spatial understanding.

Contribution

It introduces a VLM-based approach combining dual-camera perception, closed-loop action refinement, and a reflective equilibrium loop for real-world minigolf mastery.

Findings

Effective in offline inference with recorded trajectories

Combines perception, action refinement, and reflective reasoning

Demonstrates advanced embodied intelligence in real-world tasks

Abstract

Minigolf is an exemplary real-world game for examining embodied intelligence, requiring challenging spatial and kinodynamic understanding to putt the ball. Additionally, reflective reasoning is required if the feasibility of a challenge is not ensured. We introduce RoboGolf, a VLM-based framework that combines dual-camera perception with closed-loop action refinement, augmented by a reflective equilibrium loop. The core of both loops is powered by finetuned VLMs. We analyze the capabilities of the framework in an offline inference setting, relying on an extensive set of recorded trajectories. Exemplary demonstrations of the analyzed problem domain are available at https://jity16.github.io/RoboGolf/