Each worker localizes and plans its own paths, avoiding obstacles without central control.
Part Detection
Onboard perception identifies and locates parts for pickup, placement or storage.
Generalized Manipulation
A robotic arm and gripper handle common part geometries — grasping, transporting, and placing bricks with verification.
Visual Sensing
WEAVE reads the build specification visually, mapping booklet-style instructions to physical placement steps.
Task-Based Operation
Workers claim buildable steps and coordinate with minimal broadcasting between the Foreman and the swarm.
Fault Tolerance
The system degrades gracefully under partial robot failure, with stigmergic coordination filling the gaps.
Placeholder imagery — robot renders and hardware photos to follow.
System Architecture
Interactive Architecture Explorer
A zoomable map of how a single instruction propagates from the Foreman through the swarm and down to physical brick placement — the visual companion to the WEAVE paper. Hover a subsystem for a summary; click it for detail.
The worker can now show what it’s doing with a light on top — white idle, blue driving, green carrying, red fault. It runs separately from the robot’s brain, so if it crashes the robot keeps building.
The worker now builds the full three-part plate using only its onboard depth camera, with no access to the simulator’s true positions — the one shortcut that wouldn’t survive a real table.
A replicated build plan that every robot carries, with no central scheduler — claiming, recovery, and convergence proven in software.
Done
Single-worker build in simulation
One simulated worker driving the full loop in Gazebo: drive, pick, place, and self-verify each step against ground truth — placing within millimeters of target.
Next
Multi-robot decentralized build
Many workers claiming and assembling concurrently, with congestion-aware navigation and coalitions for jobs too large for one arm.
Next
Perception & part detection
Replacing simulation ground truth with real onboard vision — identifying, localizing, and verifying LEGO parts under real lighting.
Planned
Hardware bring-up
First worker chassis and gripper iterations on the 3D printer, then sim-to-real onto onboard compute.
