RoboMission Fleet — Autonomous Warehouse Simulator

Inspiration

We were inspired by real-world fulfillment centers and automated warehouses, where robots must navigate changing environments, avoid unexpected obstacles, and follow instructions from human operators. We wanted to build a system where robots could understand natural-language missions and adapt their paths dynamically when barriers appeared, making warehouse automation more flexible and intelligent.

What it does

RoboMission Fleet is a robot simulation system that translates natural-language instructions into missions and uses adaptive A* pathfinding to navigate dynamic environments. Robots can avoid newly introduced obstacles, replan routes in real time, and manage cargo while providing live telemetry to explain their decisions. An offline command parser ensures mission processing remains available even without AI connectivity.

How we built it

We built RoboMission Fleet using Vanilla TypeScript, Vite, HTML5 Canvas, and the Gemini LLM API. The system uses a four-directional Manhattan A* pathfinding algorithm to calculate efficient routes, with a heuristic of $h(n) = |x_1 - x_2| + |y_1 - y_2|$. A dual-parser system combines Gemini-based natural-language interpretation with an offline regex fallback for reliable command processing. We also developed a discrete-tick simulation and state machine to manage robot movement, cargo handling, obstacle avoidance, and real-time replanning, supported by live telemetry for explainability.

Challenges we ran into

One challenge was maintaining clear visuals across different screen resolutions. We addressed this by using a fixed logical canvas resolution with device pixel ratio scaling. Another challenge was implementing real-time path replanning when obstacles appeared during robot movement. We refined the simulation loop to pause movement, recalculate routes, and resume animation without synchronization issues. We also carefully ordered world resets and obstacle placement to ensure our failure scenarios behaved as expected.

Accomplishments that we're proud of

We're proud of combining AI-powered natural-language mission parsing with deterministic pathfinding in a responsive, interactive simulation. Our system supports real-time obstacle detection and route replanning, clear visual feedback, and an offline parsing fallback. We also achieved sub-millisecond pathfinding performance in our tests and passed all 20 Vitest unit tests, demonstrating the reliability of our core logic.

What we learned

We learned the importance of planning system architecture before developing the user interface. Separating pathfinding, command parsing, simulation, and rendering made the project easier to test and maintain. We also discovered how combining AI-powered natural-language understanding with deterministic algorithms and offline fallbacks can improve both usability and reliability. Finally, our 20/20 passing Vitest tests reinforced the value of automated testing in catching edge cases early.

What's next for RoboMission Fleet — Autonomous Warehouse Simulator

We plan to extend RoboMission Fleet to coordinate multiple robots simultaneously, optimize task allocation, and prevent collisions between robots. We also want to integrate more realistic warehouse environments, improve mission understanding, and explore connections with physical robots using ROS 2. Ultimately, our goal is to make robot fleets more adaptive, explainable, and accessible through simple natural-language commands.

Built With

• a-star-algorithm
• autonomous-vehicles
• css3
• explainable-ai
• heuristics
• html5
• javascript
• node.js
• pathfinding
• robotics
• simulation
• state-machine
• typescript
• vite
• vitest