B.O.B. (Brick On Board) is a roughly 3-foot, fully 3D-printed Lego minifigure riding a motorized skateboard base. I designed the entire robot in SolidWorks — the scaled-up minifig shell, the internal servo and electronics mounts, and the skateboard drive base — so that a toy-accurate figure could actually move: arms that raise and wave, a head that turns and nods, and a base that drives and dances. You control it by talking to it; an on-device intent classifier on an ESP32-S3 turns natural-language commands into motion.

• Full robot modeled in SolidWorks, from the scaled minifig geometry down to every internal mount
• Minifig proportions kept faithful to the real Lego figure while hiding all actuators and wiring inside the printed shell
• Shoulder joints driven by MG996R servos mounted inside the torso, sized for the torque of the long printed arms
• Neck joint driven by an SG90 servo for head turning, nodding, and shaking
• Skateboard-style base with TT gearmotors, designed to carry the figure stably while driving, turning, and dancing
• Large parts split into printable sections with integrated fastening and mounting features
• Rendered in SolidWorks, including a transparent view of the internal layout

• On-device TensorFlow Lite Micro intent classifier on the ESP32-S3, mapping commands to 14 actions (arms, head, drive, dance, stop)
• Handles synonyms, speech-to-text errors, multi-step sequences, and parameters like “turn left 45 degrees”, and rejects low-confidence predictions
• Motion planner and command queue turn intents into timed servo and motor actions without blocking
• Local YOLO vision on a Raspberry Pi answers “what do you see?” out loud, with distance estimates
• TinyStories-based storytelling, routed separately from motor commands

• Scaling a toy-sized figure up to 3 feet while keeping it light enough for the servos and stable on a moving base
• Packaging servos, controllers, and wiring inside the minifig shell without breaking its silhouette
• Giving the arms and head enough range of motion for simple gestures to read as expressive
• Fitting a useful language model onto a microcontroller and turning bare intents into natural motion

• devpost submission →
• demo video →