DuckSpace.org
A zero-installation, browser-based 3D physics test bench serving the Open Duck Mini and Microduck bipedal robot ecosystem. It integrates the MuJoCo physics engine and ONNX Runtime Web, supporting in-browser loading and testing of community-contributed locomotion policies and high-level control modules.
Timeline (last 90 days)
DuckSpace.org is a zero-install browser-based 3D physics test bench serving the Open Duck Mini / Microduck bipedal robot ecosystem.
DuckSpace.org combines the MuJoCo physics engine with ONNX Runtime Web and runs in the browser at 50Hz.
DuckSpace.org currently supports 16 .onnx locomotion policies, covering walking, sprinting, forward rolls and recovery, ground pecking, single-leg flamingo balance and other motions.
DuckSpace.org allows users to upload their own .onnx files and test them directly in the browser.
DuckSpace.org integrates several community "brain" modules: quack-nav (2.5D occupancy grid SLAM + A* planner + [351→9] neural network navigation policy + virtual stair/fall protection), DuckBuddy (natural language multi-step route execution + hot servo model), kissqy R17-V167 and fanhao375 duck-echo (Feetech HD-1910 six-policy switcher, action_scale tuning, negative safety voice guard, BPM-paced nodding).
DuckSpace.org allows users to switch servo profiles, adjust carpet thickness, and toggle ground obstacles, robot vacuums, cats and other distractors.
Whether the 50Hz operating frequency of browser-based MuJoCo and its simplified contact model can accurately reflect a policy's behavior on a physical robot is the issue professional developers care about most.
Domain randomization sliders and A/B policy comparison are the two candidate features of highest value to policy developers.
For bipedal robot RL policy developers, a truly useful browser test bench should prioritize three things: policy loading and fast replay, adjustable domain randomization parameters, and records comparing simulation versus physical performance.
All Facts (9)
DuckSpace.org is a zero-install browser-based 3D physics test bench serving the Open Duck Mini / Microduck bipedal robot ecosystem.
50%UnverifiedDuckSpace.org combines the MuJoCo physics engine with ONNX Runtime Web and runs in the browser at 50Hz.
50%UnverifiedDuckSpace.org currently supports 16 .onnx locomotion policies, covering walking, sprinting, forward rolls and recovery, ground pecking, single-leg flamingo balance and other motions.
50%UnverifiedDuckSpace.org allows users to upload their own .onnx files and test them directly in the browser.
50%UnverifiedDuckSpace.org integrates several community "brain" modules: quack-nav (2.5D occupancy grid SLAM + A* planner + [351→9] neural network navigation policy + virtual stair/fall protection), DuckBuddy (natural language multi-step route execution + hot servo model), kissqy R17-V167 and fanhao375 duck-echo (Feetech HD-1910 six-policy switcher, action_scale tuning, negative safety voice guard, BPM-paced nodding).
50%UnverifiedDuckSpace.org allows users to switch servo profiles, adjust carpet thickness, and toggle ground obstacles, robot vacuums, cats and other distractors.
50%UnverifiedWhether the 50Hz operating frequency of browser-based MuJoCo and its simplified contact model can accurately reflect a policy's behavior on a physical robot is the issue professional developers care about most.
50%UnverifiedDomain randomization sliders and A/B policy comparison are the two candidate features of highest value to policy developers.
50%UnverifiedFor bipedal robot RL policy developers, a truly useful browser test bench should prioritize three things: policy loading and fast replay, adjustable domain randomization parameters, and records comparing simulation versus physical performance.
50%