Explore / Case study / Inner Monologue — Feedback-Driven Robot Planning

Inner Monologue — Feedback-Driven Robot Planning

Demo only

Use fresh execution, scene and human feedback to revise a robot plan.

by admin · listed 5 days ago

⚡ What System1 does here

The language model updates high-level plans from object, scene, success and human feedback; existing low-level skills execute movements. The study investigates this planning approach without additional model training. No Jev integration is established.

About

A research method for feedback-driven robot planning: after an action fails, the planner can revise the next steps using updated observations. Evaluations cover simulated tabletop tasks, real tabletop manipulation and a mobile robot in a kitchen. Evidence and limits: Results depend on feedback quality and available skills. Some recovery demonstrations deliberately induce failures through human intervention. These research results do not establish reliability across arbitrary household tasks, robots or production settings. We did not run the project. Original project images and demonstrations (with attribution):

Official project demonstration ↗

Kitchen mobile robot replanning from failure feedback ↗

Sources: https://innermonologue.github.io/ https://arxiv.org/abs/2207.05608 https://youtu.be/0sJjdxn5kcI

🧪 Community test reports

0 reports · 0 passed on the real model

No one has tested this yet. Be the first.

💬 Discussions

New post →

No discussions yet.