FLA · Walking: the Hiking recipe on the Booster K1 · 16 Nov 2026 · all reports · summary

Porting a G1 walking recipe to the Booster K1

The TactileStep rebuild ran on the Unitree G1. The sensing-sole work runs on the smaller Booster K1. The first step toward tactile walking on the K1 was to train an ordinary walking policy there with the Hiking in the Wild recipe. Three attempts; none produced a policy that walks well in both simulators.

Simulation only. The goal of this stage was not reached.

In short

  1. Two attempts from scratch learned to stand, not walk: 0.01 and 0.09 m/s at a command of 0.5 m/s after 400 iterations. Reward settings sized for the larger G1 made standing still pay.
  2. A warm start from my own K1 walking policy walked. At iteration 200 it passed the Isaac Lab check: no falls in 24 runs, 0.41 m/s.
  3. The same policy failed in MuJoCo: 10 falls in 24 runs and 0.26 m/s.
  4. More training made it worse. Its speed fell from 0.41 to 0.19 m/s by iteration 1,300, as the terrain curriculum got harder and the action noise grew. Training stopped at iteration 1,478.
  5. So the tactile stage on the K1 was not started.
1

What had to change from the G1 version

Hiking in the Wild is a published recipe for training a G1 walking policy on rough terrain in Isaac Lab, with a motion prior (AMP) that rewards a human-like style. Porting it to the K1 meant changing:

2

Three attempts

Training-log values at iterations 50 / 200 / 400 to 500 / last. The test: command 0.5 m/s on flat ground, 3 seeds × 8 robots, 20 s.

Attempt 1: from scratchAttempt 2: from scratch, rescaled termsAttempt 3: warm start
Test speed (m/s) and fallsit200 0.00, it400 0.01; 0 fallsit200 0.00, it400 0.09; 0 fallsit0 0.43 (5/24 falls), it200 0.41 (0/24, pass), it500 0.28, it1300 0.19
Episode length (steps)34 / 41 / 33 / 11237 / 67 / 70 / 63140 / 619 / 580 / 769 (it1478)
Speed-tracking reward per episode0.78 / 1.50 / 1.61 / 1.470.47 / 1.06 / 1.18 / 1.191.36 / 13.5 / 6.5 / 13.3
Curriculum tracking score0.02 to 0.040.01 to 0.030.03 / 0.34 / 0.16 / 0.33
Air-time reward / standing penalty0.003 / −0.190.003 / −0.13 to −0.210.01 to 0.05 / −0.7 / −1.1 / −2.2 / −1.4
Style (AMP) reward0.32 → 0.500.30 → 0.450.30 / 0.34 / 0.37 / 0.42
Action noise (std)0.92 → 0.870.91 → 0.850.40 / 0.43 / 0.49 / 0.65
Terrain level0 → 3.80 → 3.00.08 / 0.82 / 4.17 / 3.10
Ends: trunk too low / time-out0.89 → 0.32 / 0 → 0.630.86 → 0.35 / 0 → 0.620.58 / 0.26 / 0.26 / 0.20; time-out 0.17 → 0.78

Reading the table. Attempts 1 and 2 never learned to move: short episodes, a flat tracking reward and a test speed near zero. Attempt 3 started from a policy that already walks and kept walking, with long episodes and a much higher tracking reward. Its test speed, though, peaked at iteration 200 and then fell.

3

Why the first two stood still

Attempt 1 was stopped at iteration 558. Its checkpoint at 500 moved 0.01 m/s at a command of 0.3, 0.00 at 0.5 and 0.22 at 0.8; with action noise on, 0.19 at 0.5. It did not fall. Since it did move at 0.8, the test was working; the cause lay in the port:

Attempt 2 fixed these (tracking width 0.3, thresholds × 0.6, only the catwalk clips) and still reached only 0.09 m/s at iteration 400, with the tracking reward flat at 1.06 to 1.19 per episode. After two failures from scratch, the third attempt used a warm start.

4

Why the third got worse

Read from the training numbers only, without a separate experiment: between iterations 200 and 500 the terrain level jumped from 0.8 to 4.2 (onto the low steps), the tracking reward halved (13.5 → 6.5) and the standing penalty doubled (−1.1 → −2.2). The action noise kept rising (0.40 → 0.65, driven by the entropy bonus) and the style reward kept rising (0.30 → 0.42), while the speed of the noise-free policy on flat ground fell from 0.41 to 0.19 m/s. By iteration 1,478 the tracking reward in training had recovered to 13.3: the noisy policy still tracked during training, but its mean action walked slower.

Attempt 3, iteration2005007001,0001,300
Test speed at command 0.5 (m/s)0.410.280.220.210.19

Reading the table. The best policy came early and training only took it further away. The decay coincides with both the harder terrain and the growing action noise; which of the two causes it was not separated.

5

The best checkpoint in Isaac Lab and MuJoCo

Attempt 3 at iteration 200, tested in the training simulator (Isaac Lab) and in MuJoCo with the same observations and depth pipeline (training friction 1.0, the same PD gains applied as torque control, the robot's effort limits, no armature; 3 seeds × 8 perturbed starts, 20 s). Command 0.5 m/s.

SimulatorFallsSpeed (m/s)Heel / flat / toePeak landing forceCheck
Isaac Lab, iteration 0 (the starting policy)5 / 240.430.45 / 0.54 / 0.00255 N–
Isaac Lab, iteration 2000 / 240.41 ± 0.000.23 / 0.76 / 0.01297 ± 6 N (1.54 BW)pass
MuJoCo, iteration 20010 / 240.26 ± 0.010.29 / 0.60 / 0.11262 ± 4 N (1.36 BW)fail

Reading the table. In Isaac Lab, 200 iterations of the recipe removed the starting policy's falls at about the same speed and shifted its landings from heel toward flat. In MuJoCo the same policy fell in 10 of 24 runs (5, 3 and 2 per seed) and walked at 0.26 m/s, so it does not transfer between simulators. BW is body weight. The MuJoCo replica was validated beforehand: at the first step its joint readings match Isaac Lab up to noise and its depth image is identical.

The MuJoCo falls may come from the starting policy rather than from this training: it was trained without observation noise or domain randomisation, and it has not itself been tested in MuJoCo.
6

Clips

Attempt 3, iteration 200, Isaac Lab, command 0.5 m/s: side view (left) and feet (right). The best checkpoint.
Attempt 2, iteration 400, command 0.5 m/s: the policy shuffles almost in place (0.09 m/s in the test).

Training recipe: Hiking in the Wild (CC BY-NC 4.0), used for research. The policies shown were trained on the K1 here; the pretrained G1 policy is not shown.

7

What to try next, and limits

Scores here are not F1: test speed at a command of 0.5 m/s, falls out of 24 runs (3 seeds × 8 robots, 20 s), share of heel, flat and toe landings, and peak landing force. Training recipe: Hiking in the Wild (CC BY-NC 4.0), used for research; its code, data and weights are not included. Related: TactileStep on the Unitree G1, the simulated sole up close, step 1, first simulation run.