Relaxed IK

This project is part of a Master's thesis in robotics at the National University of Singapore, School of Computing.

Thesis: Leveraging Rust for real time low computational Robotics
Author: Guilhem Mathieux (@guimath)
Advisor: Lin Shao

This repository is a fork of relaxed IK core by the University of Wisconsin-Madison's Graphics Group. It proposes some optimizations, code improvements and added functionality. It also incorporates a motion planner to have complete functionality for robotics motion control.

Introduction

The increasing complexity of modern robotic arms has led to significant increases in computational time, primarily due to the growing dimensionality of the problems they must solve. While this is manageable for applications that can pre-compute movements on powerful computers, it poses challenges for real-time applications, resulting in excessive latency.

In this thesis, we present an efficient, low-computation method for generating manipulator movements by formulating the pose generating task as an optimization problem and utilizing Rapid exploring Random Trees in order to determine a viable path to the intended goal. The entire project is implemented in Rust, leveraging its safety features and performance advantages.

Figure : example of a pre-grasp motion fully generated by this system, visualized in simulation.

Figure : example of a grasp motion fully generated by this system, run in real life.

For detailed information on the work, see the full thesis

Getting Started

Build

1. Install Rust
2. Run the interactive demo:

   make reach_full

   or

   cargo run --example reach

To compile the python wrapper use maturin:

1. start by initializing the venv:

   make init_env

   or

   python3 -m venv .venv
   . .venv/bin/activate
   pip install -U pip maturin

2. build using

   make build_python

   or

   maturin develop -F python_wrap

Config files

The config files are written in toml, and allow for robot specific modifications.

• urdf_paths: Paths relative to config file
  • robot
  • obstacles
• packages: List of package names used in urdf files (not mandatory, can help speed up urdf load)
• links:
  • base: name of the base link
  • ee: name of end-effector target
  • used_joints: list of all joints to be used
• starting_joint_values: joint values to be considered as reset. Default = 0.0 for all
• approach_dist: additional distance for pre grasp motion. Default = 0.03

For examples of config files see xarm6 file or ur5 file.

Objective weights and cost function can also be specified for example:

[objectives.x_pos]
func= {Groove={t = 0.0, d = 2, c = 0.1, f = 10.0, g = 2}}
weight= 70

An objective can be fully disabled by making the weight = 0.

For more examples and all the defaults objective configs, see the default file

Examples

• bench.rs runs a large batch of gradients calculation on the different active objectives to test time needed on each (gradient being the most computational heavy operation as it requires many objective calls for each gradient)

Objective name	batch runtime (ms)	isolated without baseline (ms)
Base line	2952	0
MatchEEPosiDoF (Z)	3812	860
MatchEEPosiDoF (X)	3876	923
MatchEEPosiDoF (Y)	3846	893
HorizontalArm	3685	733
HorizontalGripper	4870	1917
EachJointLimits (0)	3755	803
MinimizeVelocity	3555	602
MinimizeAcceleration	3762	809
MinimizeJerk	3844	892
MaximizeManipulability	23025	20072
SelfCollision (0 – 2)	5576	2623

• cost_viz.rs simple visualization tool to plot a given cost function. See the example.toml file for an example of plot with 3 different cost function. Figure: Graph produced by running cost_viz on the example.toml config.

• metric.rs Scan large area around the robot either computing IK or Motion and saves data to a bin file. May take a while for precise scan (especially for motion)

• plot_metric.rs plots a metric bin file to a heatmap. Plots can be configured by modifying the toml file produced along the bin files to change axis labels, scale range & title and more. Figure: Cost visualization heatmap of IK over the XY plane at Z=0.3m produced with metric and plot_metric.

• reach.rs Visualize robot and see motion and inverse kinematic results in simulation Figure: Screen of visualization tool when running in "full" mode.

• time_bench.rs runs a large batch of calculation on a given area and saves detailed run time for each sample in csv file. Said CSV file can be easily exploited in python (basic processing of csv and plotting in a histogram in time_bench_plot.py)
  Figure: histogram of runtime for Xarm6 produced via data from time_bench.rs and plotting from time_bench_plot.py

The metric and reach examples have some helper commands, see the Makefile