Explore original result figures, comparison tables, and robot demonstrations. Each work connects its behavior to the experiments reported in the paper.
MDOC-CBS coordinates multiple robots in a constrained scene.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: mdoc · assets/vis/gifs/MDOC-CBS/random_MDOC-CBS_10.gif
An arm executes obstacle-avoidance motion in the D3IL environment.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: mdcoas_demo · d3il_env.gif
Governed quadruped motion follows discrete footholds.
Original source timing preserved; 1× playback, looped and silent. Source animation timing is not a planning-latency benchmark.
Source: project · results/quadruped/stepping_stones_2d/deploy/governed/twogo/level_5/seed_2/governed/trajectory_mujoco.gif
Body posture and clearance adapt through a constrained corridor.
Original source timing preserved; 1× playback, looped and silent. Source animation timing is not a planning-latency benchmark.
Source: project · results/humanoid/corridor_2d/deploy/governed/twogo_zone_c/level_1/seed_2/trajectory_mujoco.gif
A rigid arm scans a surface with changing material compliance.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: mga_supplement · gifs/mga_surface_hybrid_stripes.gif
Motion and impedance adapt during peg insertion.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: mga_supplement · gifs/mga_peg_id_wide.gif
Single-robot diffusion planning in a narrow scene.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: mdoc · assets/vis/gifs/MDOC/narrow_MDOC.gif
The diffusion process on the paper L10 / seed 8 planar scene.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: project · results/single2d/mdcoas/level_10/seed_8/diffusion_steps/diffusion_steps.gif
Candidate trajectories from the same L10 / seed 8 planning case.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: project · results/single2d/mdcoas/level_10/seed_8/trajectory/trajectory_modes.gif
TCP projection of diffusion on L1 / seed 0; separate from the D3IL execution clip.
Source excerpt rescaled to 6.0 seconds; looped, silent; not real-time execution or latency evidence.
Source: project · results/d3il_avoiding/mdcoas/level_1/seed_0/diffusion_steps/diffusion_steps.gif
Real Unitree G1 navigates motion-capture-registered AR virtual obstacles.
Original hardware video, 92–112 s excerpt at 1× playback; full frame preserved, looped and silent.
Real G1 with motion-capture-registered AR virtual obstacles. This is qualitative deployment evidence.
Source: ral_submission · video/2go.mp4
Original MGA hardware footage of contact with compliant material.
Original source-video timing preserved; looped and silent. No additional speed change is applied.
MGA model-based refinement component on Dobot Nova 5, without an RL proposal prior; position commands with force feedback.
Source: mga_supplement · videos/deployment.mp4
Original MGA hardware footage of curved-surface contact.
Original source-video timing preserved; looped and silent. No additional speed change is applied.
MGA model-based refinement component on Dobot Nova 5, without an RL proposal prior; position commands with force feedback.
Source: mga_supplement · videos/deployment.mp4
Original MGA hardware footage of peg insertion.
Original source-video timing preserved; looped and silent. No additional speed change is applied.
MGA model-based refinement component on Dobot Nova 5, without an RL proposal prior; position commands with force feedback.
Source: mga_supplement · videos/deployment.mp4
Unitree H1 pushes a box forward from a fixed stance.
Original GIF timing preserved, including the final 800 ms hold; looped and silent. Repeated frames preserve source timestamps at 50 fps. Source animation timing is not a physical real-time benchmark.
MGA Unitree H1 simulation from the paper supplement. Humanoid pushing remains deferred in the current framework release; this clip does not establish a qualified V1 reproduction path.
Source: mga_supplement · gifs/mga_humanoid_fixed_stance_push.gif
Unitree H1 regulates contact force against a stationary box.
Original GIF timing preserved, including the final 800 ms hold; looped and silent. Repeated frames preserve source timestamps at 50 fps. Source animation timing is not a physical real-time benchmark.
MGA Unitree H1 simulation from the paper supplement. Humanoid pushing remains deferred in the current framework release; this clip does not establish a qualified V1 reproduction path.
Source: mga_supplement · gifs/mga_humanoid_force_regulation.gif
Unitree H1 adjusts contact to free a box from a narrow passage.
Original GIF timing preserved, including the final 800 ms hold; looped and silent. Repeated frames preserve source timestamps at 50 fps. Source animation timing is not a physical real-time benchmark.
MGA Unitree H1 simulation from the paper supplement. Humanoid pushing remains deferred in the current framework release; this clip does not establish a qualified V1 reproduction path.
Source: mga_supplement · gifs/mga_humanoid_unjamming.gif
