FLA step 1 · Booster K1 · 15 Nov 2026 · step 2: Vision from Touch · all reports

Terrain from Touch

Can a sensing sole tell a humanoid what it is standing on? A classifier trained and tested in simulation on the K1, with flat-foot walking and human-like heel-to-toe walking.

Simulation only. Nothing on this page comes from the real sole yet.

In short

  1. It works in simulation, and more sensor data helps a bit. Ground type: 0.89 with the sole, 0.82 from joints and IMU alone.
  2. The sole alone is almost as good as sole plus joints. Pressure map and shear with no joints at all: 0.88.
  3. The larger gain is support, not ground type. Knowing if a foot is in the air, on an edge or fully down: 0.93 against 0.76. A lifted robot is noticed in 0.1 s instead of 0.3 s.
  4. A sole-only model still works on a gait it never saw. A model that also reads joints does not: 0.53 against 0.07 on the human-like gait.
  5. Shear adds little on a known gait and more on a new one. +1 to +4 points on a known gait, +6 to +12 on an unseen gait. The second number is still being re-checked.
  6. The sole helps more on heel-to-toe walking than on flat-foot walking. Earlier papers mostly used flat-foot walking, based on the reviewed work.
  7. This will not transfer to the real robot as it is. The pressure simulation is roughly right, the shear simulation is weak, and sand, mud or grass cannot be simulated. A real recording is needed.
1

Experimental setup

What was compared

The same model was trained on different inputs, with and without the robot's own joints and IMU:

The six surfaces, flat-foot gait. First version of the sole (eight zones per foot).
The simulated sole on a test rig: the foot is pressed down. Color is pressure, arrows are shear. Front and back load, the middle stays empty.
2

Results by input

Walking, on surfaces and gaits the model was trained on. Scores are F1 from 0 to 1, higher is better.

InputGround typeFoot support
Joints + IMU only0.820.76
Joints + IMU + 4 pressure blocks0.830.81
Joints + IMU + 4 pressure blocks + shear0.840.81
Joints + IMU + pressure map0.880.93
Joints + IMU + pressure map + shear0.890.94
Sole only: 4 pressure blocks0.780.79
Sole only: 4 pressure blocks + shear0.820.80
Sole only: pressure map0.860.92
Sole only: pressure map + shear0.880.93
Sole only: shear, no pressure0.790.75

Reading the table. Adding the sole helps, and the pressure map helps more than coarse blocks. On ground type the gain over joints and IMU is modest: 7 points at best. The sole without any joints (0.88) is almost as good as everything together (0.89). Shear alone is about as good as joints alone.

The first version of the sole had eight zones per foot and gave a similar picture: 0.80 without the sole, 0.86 with pressure, 0.86 with pressure and shear.

3

Support, lifting and edges

This is where the sole matters most, and it matches the two failure videos: stepping off the stage, and being lifted while the policy keeps "walking".

What is measuredJoints + IMU+ 4 blocks+ pressure map
Foot support state (F1)0.760.810.93
"Foot is only partly supported" (F1)0.360.470.80
Robot lifted: time to notice0.30 s0.10 s
Foot on an edge for 0.3 s or longer: missed37 %27 %2 to 3 %
Foot on an edge for less than 0.3 s: missed74 to 85 % for every input
Lifted mid-walk at about 4 s. Forces drop to zero and the model reports both feet in the air.
Planks with gaps: feet land partly on edges.
A failure case: an edge contact the model misses.
Test rig: the foot is set down half on an edge. Only the supported part of the sole reads pressure.
4

A gait the model has not seen

Each model was trained on two gaits and tested on the third. Models that read joints learn the gait as much as the ground, so they break. A model that reads only the sole keeps part of its skill.

Ground type F1 on an unseen gaitSole onlySole + joints
Pressure map + shear, average of the three gaits0.630.53
Pressure map, average0.570.51
4 blocks + shear, average0.520.37
4 blocks, average0.400.35
Pressure map + shear, tested on the heel-to-toe gait0.530.07

What follows. The same models score 0.78 to 0.89 on a gait they were trained on, so a new gait costs a lot in every case. The classifier has to see the controller that will run on the real robot, or it should read the sole only.

5

Flat-foot against heel-to-toe

The robot walks about a minute across a course with all six surfaces in random order. The score is the share of time the ground type is named correctly.

GaitJoints + IMUWith the soleGain
Flat-foot policy G1130.850.88+3
Flat-foot policy G800.800.84+4
Heel-to-toe (SONIC)0.700.81+11
All three0.790.84+6

What follows. On the human-like gait joints alone say less about the ground, so the sole adds more. This is one result from one set of runs, not yet a separate confirmed test.

In the videos, the strips under the picture fill in as the robot walks: the true ground, then the model with the sole, then joints and IMU only, then a hand-written detector with no learning (0.31, shown only for reference). The panel on the right is the simulated sole.

Mixed course, flat-foot policy G113.
Mixed course, flat-foot policy G80.
Mixed course, heel-to-toe gait (SONIC).
Grippy floor, flat-foot gait.
Grippy floor, heel-to-toe gait.
Slippery floor, flat-foot gait.
Slippery floor, heel-to-toe gait.
The walking is not robust. Every part of the course was crossed in 19 of 40 runs by G113, 22 of 40 by G80 and 6 of 40 by SONIC. Runs were scored up to the fall. The three videos are runs without a fall.
6

Shear

Sole-only models, ground type F1Without shearWith shearGain
Pressure map, known gait0.860.88+2.5
4 blocks, known gait0.780.82+4
Pressure map, unseen gait (average)0.570.63+6
4 blocks, unseen gait (average)0.400.52+12
Pressure map, unseen heel-to-toe gait0.420.53+11
4 blocks, unseen heel-to-toe gait0.240.36+12
Test rig: foot pressed down and dragged sideways until it slides, slippery floor (friction 0.3). The four arrows are shear per block.
The same on a grippy floor (friction 0.9).
Test rig: the foot is twisted in place.
Walking from a grippy strip onto a slippery one.
Close-up of the sole while walking on a grippy floor, slowed to 0.25x.
Close-up on a slippery floor, 0.25x.
Close-up on a slope, 0.25x.

The close-ups are from the flat-foot gait. Close-ups for the heel-to-toe gait are not rendered yet.

7

How many points are needed

Trained on the standard sole, tested onPressure mapMap + shear4 blocks4 blocks + shear
the same sole0.860.880.780.81
shifted sensor settings (noise, gain, range)0.530.600.690.74
a different contact model for the sole0.210.230.590.59
a foot-shaped point layout instead of a rectangle0.240.210.760.74

Ground type F1, models with joints and IMU. Values in the lower three rows are read from the chart and rounded. Chart: v2_fig5_sensitivity.png.

8

Limits of the simulation

10

Real-world validation requirements

  1. PaXini documentation, SDK or simulation examples: pressure and shear channels, point layout, ranges, and adapter requirements.
  2. A 15 to 30 minute K1 recording: standing, manual loading, and walking on two or three surfaces.
  3. The walking controller used during recording, to account for gait dependence.
11

Next

12

All other clips

Soft ground, flat-foot gait.
Rough ground, flat-foot gait.
Slope, flat-foot gait.

Numbers are from the second simulation run unless stated: test split, walking, mean of three training runs. Full reports: REPORT_V2.md, REPORT.md. All tables: tables_v2.md. Charts: inputs, partial support, unseen gait, course.