Probabilistic Contact Estimation

This ROS package contains:

• A real-time contact detection module that operates in real-time(~500hz) and uses only IMU mounted on the foot of the robot. It has been tested on a quadrupedal GO1, a real TALOS humanoid robot and a simulated ATLAS humanoid.
• Datasets (./offline/data/) from an ATLAS simulated humanoid in RAISIM and a real GO1 quadruped on various terrains
• An offline contact probability estimator module to test on custom datasets.

The published manuscript can be found at: https://ieeexplore.ieee.org/document/10161485

Setup

The only dependency for the offline module is sklearn. Run:

$ pip install -U scikit-learn

Then define the data filename you want to use at offline/src/filename.py and:

$ ./atlas_contact_estimation.py
             or 
$ ./go1_contact_estimation.py

Please note in case you are using another robot, you will need to fine-tune the threshold parameters inside the file by extracting them from a normal gait pattern (no slip). These thresholds are robot/control specific.(More details about how to extract them will be added soon)

Datasets

ATLAS bipedal

All files at ./offline/data/ATLAS_ have the same structure:
Fx | Fy | Fz | Tx | Ty | Tz | ax | ay | az | wx | wy | wz | label

The refresh rate of the meausurements is 1000 hz

Description

• ATLAS_01ground.csv : 0.1 static friction coef., ATLAS walking around.
• ATLAS_01ground_3steps.csv: 0.1 static friction, 3 steps of ATLAS
• ATLAS_01ground_003slip.csv: 0.1 ground static friction, 0.03 static friction on extremely slippery surface.

GO1 quadrupedal

All files at ./offline/data/GO1_ have the same structure:
| Fz | ax | ay | az | wx | wy | wz |

The refresh rate is 250 hz for the IMU and 500 hz for the Fz

• GO1_matress.csv: Walking on soft terrain (a matress)
• GO1_normal_surface.csv: Walking on a normal surface.(these data are unsynchronized)
• GO1_slippery_surface.csv: Walking on a low friction surface. A greased smooth surface, extremely slippery.

Real-time Probabilistic Contact State Estimation

Using Unitree's Go1 quadrupedal robot and low-end IMU sensor (Arduino RP2040 integrated IMU) mounted on the foot.

Description

An analytical package description of the real experiment with the Go1 robotic dog of Unitree. This package is developed to extra support the theoretical analysis and the simulated experiments of the submitted paper to ICRA 2022 with title "Probabilistic Contact State Estimation for Legged Robots using Inertial Information".

Dependencies

Unitree's Go1 legged robot:

• unitree_ros_to_real
• unitree_ros
• unitree_legged_sdk

Arduino Nano RP2040's IMU:

• rosserial_python

System

• Ubuntu 20.04
• ROS Noetic

Installing

• Follow the instructions of the above listed depended packages.
• Under your 'workspace/src', git clone ProbabilisticContactEstimation package.

git clone https://github.com/MichaelMarav/ProbabilisticContactEstimation

Rename the package to "pce" (Compatible name for catkin package)

Executing program

Terminal 1

Connect to the real Go1 robot

sudo pce/BashScripts/ipconfig.sh

roslaunch unitree_legged_real real.launch

Terminal 2

Arduino pubs IMU data at \imu

sudo pce/BashScripts/imu.sh

Terminal 3

Initialize IMU bias info

• Set your path of "/.. /src/pce/src/imuBias.txt", at line 137 of 'init_imu_force.cpp' .

rosrun pce init_imu_force

rosrun pce slip_recovery

Experiment setup

Mount IMU on a leg (or multiple legs) of Go1 Unitree's. Set the communication between Arduino Nano PR2040 and your Laptop through USB - micro cable.

Citation

If you are using this work please use the following citation:

@INPROCEEDINGS{10161485,

  author={Maravgakis, Michael and Argiropoulos, Despina-Ekaterini and Piperakis, Stylianos and Trahanias, Panos},

  booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)}, 

  title={Probabilistic Contact State Estimation for Legged Robots using Inertial Information}, 

  year={2023},

  volume={},

  number={},

  pages={12163-12169},

  doi={10.1109/ICRA48891.2023.10161485}}