MAJKA

Simulation autonomy update

MAJKA Quadruped

A small quadruped robot stack focused on simulation-first locomotion, obstacle avoidance, and turning behavior.

Current result Video file

Latest recorded run

Slalom without contacts

The current demo combines a trained walking policy with a separate autonomy layer that chooses steering and speed from proximity-style sensor input.

3.35 m distance covered
18.8 s recorded run time
0 collision steps
21 cm minimum obstacle clearance

System shape

Two layers, one demo path

01

Gait policy

Residual walking model keeps the robot moving in a stable enough style for obstacle experiments.

02

Autonomy policy

Planner reads simplified distance sensors and changes speed or heading before contact.

03

Demo logger

Runs are exported as video and CSV traces so each behavior can be compared across iterations.

Simulation clips

Current behaviors

Two separate checks are useful right now: obstacle navigation for the full autonomy stack and an isolated rotation clip for turn-in-place behavior.

Autonomy

Slalom with sensor overlay

Planner follows the course, keeps clearance from obstacles, and logs the path trail.

Locomotion

Turn-in-place check

Separate rotation behavior for cases where the robot should reorient before moving forward.

Simulation roadmap

What to improve next

  1. Done Walking and slalom demo

    Recorded visual demo with sensor overlay, obstacle clearance, and no collision steps.

  2. Done Turn-in-place clip

    Separate rotation behavior is visible enough to compare against forward-only obstacle recovery.

  3. Now Cleaner behavior switching

    Blend walk, stop, pivot, and resume so the robot chooses a turn before pushing into close obstacles.

  4. Next Harder procedural courses

    Randomize blocks, corridors, starts, and goal angles to make the policy less tuned to one showcase path.