RoboBrain helps student robotics teams turn scattered engineering memory into clear action. Robotics teams do not only build a robot. They manage code, mechanical logs, meeting decisions, test reports, rule notes, task lists, and strategy across different tools and people. When a new or non-programming member joins, they often see the robot but cannot easily understand why decisions were made, what failed before, or what they should do first.
Our prototype starts with a VEX-style robotics team because we have direct robotics competition experience, but the idea can extend to RoboMaster, FRC, FTC, and other student engineering competitions. The demo loads a synthetic team knowledge base containing engineering logs, a code README, autonomous test reports, meeting notes, mock commit summaries, rule notes, and onboarding tasks.
RoboBrain generates three main outputs. First, it creates a project snapshot that explains the robot's current state, key risks, and next action. Second, it builds a first-week roadmap for a new member so they can move from confusion to a useful contribution. Third, it answers team questions with retrieved sources, so a user can ask things like "Why did autonomous fail last time, and what should I do first?"
We built the MVP as a lightweight web app using HTML, CSS, and JavaScript. The current version demonstrates the AI workflow locally with retrieval and templated generation over synthetic data. A production version would connect to authorized team files such as GitHub commits, engineering notebooks, meeting notes, and competition rule documents, then use an LLM with retrieval-augmented generation.
The main challenge was narrowing the idea. A robotics "second brain" can become too broad, so we focused the MVP on one painful moment: helping a new member understand the current build and take a safe first action. We are proud that the prototype keeps humans in control by showing sources, uncertainty, and review reminders instead of pretending the AI can make final robot decisions.
Next, we would add real integrations for GitHub, team documents, and rule manuals, plus permissions so teams can decide what information can be used by the assistant.
Built With
- css
- decision-support
- html
- javascript
- natural-language-processing
- responsible-ai
- retrieval-augmented-generation
- synthetic-data
Log in or sign up for Devpost to join the conversation.