idea-discovery-robot

Orchestrates robotics literature survey, idea generation, novelty check, and review into benchmark-grounded research ideas.

Updated Jul 9, 2026
One-click install
npx skills add https://github.com/Lingjie-wang/autoRL --skill idea-discovery-robot-lingjie-wang
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: idea-discovery-robot
Source: https://github.com/Lingjie-wang/autoRL/tree/main/Auto-claude-code-research-in-sleep/skills/idea-discovery-robot
Command: npx skills add https://github.com/Lingjie-wang/autoRL --skill idea-discovery-robot-lingjie-wang

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a broad robotics direction into a concrete, publishable research idea is hard: generic idea generators ignore embodiment, benchmarks, sim2real constraints, and evaluation quality. This Skill runs a structured pipeline that grounds every idea in a specific robot, benchmark, and sim-first pilot plan. ## Core Features & Use Cases - Robotics-Aware Literature Survey: Groups prior work by embodiment, task family, benchmark, action interface, and sim2real setup, targeting CoRL, RSS, ICRA, IROS, and RA-L venues. - Filtered Idea Generation: Produces candidate ideas with target benchmark, bottleneck, minimum sim-first pilot, mandatory metrics, and expected failure modes, while downranking demo-driven or hardware-dependent ideas. - Novelty Verification and External Review: Runs per-idea novelty checks and a simulated CoRL/RSS/ICRA reviewer pass, then writes a structured IDEA_REPORT.md with an evidence package. - Use Case: A researcher exploring bimanual manipulation runs the pipeline and receives three ranked, benchmark-grounded ideas with sim pilot plans, baselines, failure metrics, and reviewer feedback—without touching real hardware. ## Quick Start Ask the assistant to run robotics idea discovery for a direction such as "drone navigation" or "bimanual manipulation", optionally adding constraints like sim-only pilots or a target venue.

Frequently Asked Questions about idea-discovery-robot

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate robotics research ideas for a specific direction?

Invoke the skill with a direction like "bimanual manipulation" or "drone navigation". It frames the problem by embodiment, task family, and benchmark, surveys recent CoRL/RSS/ICRA work, then generates and filters ideas with sim-first pilot plans.

How does the pipeline check whether a robotics idea is novel?

Each top idea goes through a novelty-check phase that includes embodiment, task family, benchmark, sensor stack, controller type, and sim2real angle. It flags ideas that are just old methods on new benchmarks or VLA/VLM applications without new mechanisms.

Does the workflow require access to a real robot?

No. The pipeline is simulation-first by default and never auto-proceeds to physical robot testing. Ideas needing hardware are marked as needing physical validation and require explicit user approval before any real-robot step.

What benchmarks and simulators does the idea discovery consider?

The survey classifies work by benchmarks such as ManiSkill, RLBench, Isaac Lab, Habitat, Meta-World, CALVIN, and LIBERO. Ideas without a concrete simulator, benchmark, or credible baseline are rejected or heavily downranked.

What happens if no simulator or benchmark is available for my idea?

The workflow produces a concrete pilot plan instead of forcing execution and labels the idea high execution risk. It stops at the report and asks whether to build infrastructure or pivot to a more executable idea.

What output does the robotics idea discovery pipeline produce?

It writes a structured IDEA_REPORT.md containing the robotics problem frame, landscape matrix, ranked ideas with risk levels, eliminated ideas with reasons, and the evidence package of baselines, metrics, and failure cases for the top idea.