MotrixArena S1 Simulated RL Challenge
Teams trained quadruped robot dogs on Motphys's in-house MotrixLab platform to clear a four-stage, multi-terrain obstacle course — 151 teams competed this season.
Through reinforcement learning, quadruped robots cross any terrain autonomously
MotrixArena S1 was hosted by Motphys, run by the Xbotics embodied-AI community, and co-organized by Vbot and D-Robotics. Participants began with flat-ground navigation and worked up to multi-terrain obstacle crossing, learning to control quadruped robot dogs across any terrain by writing the code themselves — all on Motphys's in-house MotrixLab training platform.
Two-stage tasks
From flat-ground navigation to multi-terrain obstacle crossing — difficulty escalates stage by stage, covering quadruped control from basics to full terrain.
Multi-terrain · domain randomization
Bumps, stairs, suspension bridges, and randomized ground parameters systematically test policy robustness and generalization.
Full-stack in-house platform
MotrixLab training framework + MotrixSim physics engine, with PPO built in — large-scale parallel training, open-source and free.
Two-stage challenge · tasks & format
STAGE 1
Flat-ground navigation
The entry-level stage: on flat ground, the quadruped walks steadily from the start to a target point — the focus is basic navigation and walking control.
STAGE 2
Quadruped obstacle course
From start to finish across three sections of escalating difficulty, comprehensively testing the quadruped's all-terrain control and algorithm tuning.
Bumps & slopes
Rough terrain + slopes
Stairs & bridge
Wave terrain + stairs + suspension bridge
Random ground
Randomized ground parameters (bumps, friction, etc.)
Results · champion & runner-up spotlight
151 teams registered and competed fiercely in simulation. The total score sums flat-ground navigation plus the three obstacle sections — here are the two teams at the top.
🏆 Champion · First Prize
scav
TU Berlin
Total · #1
scav · 1/2
20
Flat nav
18
Bumps & slopes
55
Stairs & bridge
15
Random ground
- ▸The only team to clear all three obstacle stages with a single policy model
- ▸The only run to double back from the bridge to the riverbed for the red-packet bonus — zero falls, remarkably smooth
🥈 Runner-up · Second Prize
just do it
UCAS
Total · #2
just do it · 1/2
20
Flat nav
20
Bumps & slopes
35
Stairs & bridge
20
Random ground
- ▸Staged training plus reward reshaping — steadily clearing every terrain.
- ▸Switching from another framework to MotrixLab, the team found it “lighter to pick up and smoother to run”.
The platform
Start training on MotrixLab
The platform MotrixArena teams used — open-source, free, and ready to go.
Explore MotrixLab