3axis
A simulated humanoid robot walking across a cracked desert hardpan at dusk. A white arrow under one foot shows the normal force, a smaller arrow shows the shear force along the terrain.

the human layer forrobot locomotion

Robots slip on terrain that people cross without incident, and we measured the required-friction gap that explains why.

recorded simulation

surface
normal force
0.00BW
shear force
0.00BW

1 / 4 · one terrain

How robots learn to walk today.

Most locomotion policies are trained and controlled against a single terrain, at one friction coefficient assumed in advance.

2 / 4 · recorded on ice

Change the terrain.

That assumption fails once the terrain changes: the robot either slips or stays upright by loading the surface harder to compensate.

recorded simulation
robots slip
recorded simulation
stays upright

3 / 4 · three axes

What three axes see.

We record the normal force and shear force of every stance, the full mechanical demand each step places on the terrain.

4 / 4 · the missing dataset

Humans already walk on all of it.

We measure the required friction of every stance across a range of terrains, the dataset current robot training pipelines are missing.

data

48,000 stances analyzed.

A stance is one foot's contact with the terrain, strike to lift, recorded from 22 people on ramps, stairs, and treadmills.

The first test of human required-friction data in robot training.

Required friction, RCOF, is the shear force of a stance divided by its normal force.

U.S. provisional patent 64/132,415.

The sole measures all three axes of force at each stance, producing a record for every step.

what every stance demands
last stance · required friction 0.00
Recorded from real people walking: the top trace is the normal force, the bottom is the shear force, and the orange number is the required friction of each stance.
peak normal force per stance

recorded simulation

On ice the robot stays up by loading the terrain harder. Forces are in BW, multiples of the robot's body weight.

prototypes

The wearable that records required friction.

The provisional patent has been filed, and the first hardware build is in progress.

12a 12b 12c 12d 14 24 16 18 20 22 10 Fz Fx Fy SURFACE LABEL FIG. 1

the sole

38 36 34 12 32 30 FIG. 2

the layers

38 36 34 32 30 34a 34b G Fz Fx,y Fx,y FIG. 3

the force path

about

Two undergraduates at the University of Michigan.

Aditya Hansoty

B.S. Biomedical Engineering

Andrew Wong

B.S. Neuroscience + Data Science

contact

Talk to us.

hello@3axis.tech

Manuscripts available on request.