# RAPID: Robot Agentic Programming from Demonstrations

Source: [arXiv](https://arxiv.org/abs/2609.30249v1)  
Feed7 permalink: https://feed7.dev/p/2609-30249v1-1y1rmv8  
Published: 2026-09-24T17:58:21.000Z  
Trust: Needs Review (needs_review)

## Why Included

RAPID turns one visual demonstration into an executable, testable robot program, showing how agent loops can derive their own specifications, tools, and verification environment.

## Source Summary

RAPID infers a task specification, robot action primitives, and an execution environment from **one visual demonstration**, then generates, verifies, and refines a program.

## Practical Implication

The reusable pattern for coding-agent builders is the closed loop: derive executable tests and tools from an example, run the program, and refine it against observed behavior. Its object-centric representation also separates strategy from scene-specific geometry.

## Agent-Ready Context

RAPID infers a task specification, robot action primitives, and an execution environment from **one visual demonstration**, then generates, verifies, and refines a program.

The reusable pattern for coding-agent builders is the closed loop: derive executable tests and tools from an example, run the program, and refine it against observed behavior. Its object-centric representation also separates strategy from scene-specific geometry.

The evidence covers simulation and a Franka arm across **eight contact-rich tasks**, plus LIBERO-Pro evaluations. The abstract reports generalization across several variations but provides no comparative rates.

## Connected Context

Feed7 judgment across 875 accumulated Signals:

RAPID extends example-driven agent programming into physical control: one demonstration can seed not only a program but its task specification, action vocabulary, execution environment, and verification loop. The object-centric split between strategy and geometry provides a concrete route to reuse across scene variations, though eight tasks and unquantified comparative results leave the breadth of that reuse unsettled.

- [Show-Harness: Just a VLM Agent Can Play Robots](https://feed7.dev/p/2609-10522v1-1lobtfw) — Both isolate model intent from robot-specific execution through a compact action layer; RAPID adds a method for inferring that layer and its environment from a single demonstration.
- [Agents Without Code: Skills, YAML, and Filesystems Replaced Python — Philipp Schmid, Google DeepMind](https://feed7.dev/p/agents-without-code-skills-yaml-and-filesystems-replaced-python-philipp-0t7a4s8) — RAPID reinforces the idea that agent definitions can be reduced around reusable tools, while also confirming that generated programs still need an external execution-and-evaluation loop.
- [From coding to Knowledge work agents — Karan Vaidya, Composio](https://feed7.dev/p/from-coding-to-knowledge-work-agents-karan-vaidya-composio-14b5s5w) — Its generate, verify, and refine cycle supplies a concrete embodied instance of the broader requirement for verification and preflight execution around consequential tool use.

## Context Map

- Layer: agent
- Domains: coding, image
- Topics: coding-agents, harness-engineering, tool-use

## Uncertainty

- The evidence covers simulation and a Franka arm across **eight contact-rich tasks**, plus LIBERO-Pro evaluations. The abstract reports generalization across several variations but provides no comparative rates.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.