Locomotion policies that survive outside the lab.

Flat-floor walking is a demo. Kyros randomizes terrain, load, motor dynamics, latency, pushes, and partial failures until a policy can recover in the conditions it will meet in the field.

terrain_recovery_v4.0disturbance curriculum live

locomotion simterrain probe

slopeimpactrecovery

physics

policy

transfer

terrain generators across indoor, outdoor, and industrial surfaces

disturbance classes from push recovery to dropped payloads

policy export path from simulated gait to hardware trial

Bipedal gait library

Reference walking, running, side-stepping and turning gaits, plus a curriculum to fine-tune each onto your robot.

Quadrupedal terrain suite

Stairs, slopes, gravel, ice, debris. Procedurally generated terrain with full physics tuning.

Outdoor & natural terrain

Dirt, mud, grass, rock — non-flat ground with realistic friction and deformation models.

Disturbance randomization

Random pushes, dropped payloads, leg failures. Train recovery into the policy, not on top of it.

Dynamics randomization

Mass, inertia, motor torque, latency, sensor noise. Robust by construction.

Navigation-conditioned policies

Couple low-level locomotion with high-level navigation goals using the same simulation backbone.

Train a policy that walks in the wild.

Tackle procedural terrain randomization, dynamic disturbances, and leg recovery out of the box.