From dc096b8e7d3210fb53ce8f137635e358151c1319 Mon Sep 17 00:00:00 2001 From: Gabe Margolis Date: Fri, 2 Dec 2022 13:00:52 -0500 Subject: [PATCH] Update README.md --- README.md | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-) diff --git a/README.md b/README.md index 39f80f3..2e75816 100755 --- a/README.md +++ b/README.md @@ -16,7 +16,6 @@ 2. [Deploying a Custom Model](#configuration) 4. [Deployment and Logging](#deployment) 5. [Analyzing Real-world Performance](#plotting) -5. [Acknowledgements](#acknowledgements) ## Overview @@ -25,7 +24,7 @@ This repository provides an implementation of the paper: - Walk these Ways: Learning with Parametric Auxiliary Rewards for an Uncertain Future + Walk these Ways: Tuning Robot Control for Generalization with Multiplicity of Behavior
Gabriel B. Margolis and Pulkit Agrawal @@ -34,7 +33,7 @@ This repository provides an implementation of the paper:
paper / bibtex / - project page + project page
@@ -58,8 +57,6 @@ Our initial release provides the following features: **Hardware Deployment**: We provide deployment code for the Unitree Go1 Edu robot. This relatively low-cost, commercially available quadruped can be purchased here: https://shop.unitree.com/. You will need the Edu version of the robot to run and customize your locomotion controller. -Users are advised to follow Unitree's recommendations for safety while using the Go1 in low-level control mode. This means hanging up the robot and keeping it away from people and obstacles. Use our code at your own risk; we do not take responsibility for any damage. - ## Training a Model ### Installation @@ -175,6 +172,16 @@ If it does not appear, and you're working in docker, make sure you haven't forgo ## Deploying a Model + +### Safety Recommendations + +Users are advised to follow Unitree's recommendations for safety while using the Go1 in low-level control mode. +- This means hanging up the robot and keeping it away from people and obstacles. +- In practice, the main safety consideration we've found important has been not plug anything into the robot's back (ethernet cable, USB) during the initial calibration or when testing a new policy because it can hurt the robot in case of a fall. +- Our code implements the safety layer from Unitree's `unitree_legged_sdk` with PowerProtect level 9. This will cut off power to the motors if the joint torque is too high (could happen sometimes during fast running) +- This is research code; use at your own risk; we do not take responsibility for any damage. + + ### Installing the Deployment Utility The first step is to connect your development machine to the robot using ethernet. You should ping the robot to verify the connection: `ping 192.168.123.15` should return `x packets transmitted, x received, 0% packet loss`.