Loading DÆMON…
physics engine (11 MB)
0.00 s
speed0.00 m/s
falls0
bumps0
best—
level#1
next—
autopilot—
position error—
brainrun4
camerachase

Drive DÆMON

A real robot brain on real LiDAR of Pioneer Courthouse Square, Portland. Hold your phone comfortably, tilt away to drive, toward you to reverse, and turn it like a steering wheel to steer.

Simulation. LiDAR and imagery: USGS 3DEP, USGS NAIP. Streets: City of Portland.

⌨️ keyboard
🍬 0
DÆMON · trained wheel-leg robot · MuJoCo sim · real Portland city data
webcam (LifeCam HD-3000, 61°) ↓
Controller: left stick = drive · right stick = turn (or left stick sideways) · RT = boost · D-pad ↑↓ = ride height · right stick ↑↓ = lean · B = get up · A = restart · Y = camera (chase / first person / top) · Start = new maze · Back = switch level (maze / Portland) · X = show / hide the planned route · LB = sensors (lidar + webcam)
Keyboard: W/S = drive · A/D = turn · Shift = boost · Q/E = ride height · Space = get up · R = restart · C = camera · N = new maze · L = switch level (maze / Portland) · V = show / hide the planned route · I = sensors (lidar + webcam; green dots = lidar hits) · H = hide help · P = autopilot (in Portland it works out where it is on its own, routes around what its lidar + depth camera see, and finds curbs with the camera; with I on: pink = its plan, orange = things in the way, red = curbs the camera found, cyan ring = where it thinks it is, yellow = last GPS fix)
The robot balances itself with the same neural-net controller trained for the real build (MuJoCo physics, 40 kg). You only ask for speed and turn. Reach the gold ring. Portland level: real lidar ground, curbs, ramps and buildings around Pioneer Courthouse Square, plus café tables, people, dogs and pigeons in the Square, shuffled every time you load or restart (R). Add ?crowd=N to the address to keep one layout, ?crowd=0 for an empty Square.