Project Work: Application of Machine Learning and Inverse Kinematic Modelling for Avoiding Singularity in Serial Robots (In progress)

Project Description:

The project aims to study and solve a problem in universal robots called "singularity." The main goal is to create a plan for avoiding these singularity issues by using Jacobian calculations and understanding the limits of the robot's joints. This involves students working on both aspects of simulations and experiments.

In the world of serial robotics, facing problems in reaching a certain position usually happens because of three main reasons: either the robot physically can't get to the spot, its joints can't move in the right way, or there are limits on how it can move in a straight line in its workspace. This project aims to explore the workspace, find singularities, fix them with smart algorithms, adjust the paths the robot takes, and inform users if some points are impossible to reach. Additionally, the goal is to provide users with alternative ways to reach their desired destinations by making slight changes in the robot's movement direction or starting from a different spot. In other words, once singularities are detected, bypassing them can be achieved through trajectory modification or robot reconfiguration. This can be accomplished by introducing additional intermediate waypoints to guide the robot through safer joint configurations, implementing damping or smoothing techniques to stabilize motion near singularities, or dynamically reconfiguring the robot's configuration, such as adjusting joint angles, velocities, or accelerations. Moreover, limiting joint velocities or accelerations near singularity points is essential. Making a proper decision is one of the main tasks of the proposed algorithm solution.

All alerts and suggestions will appear in a VR headset to guide users. Therefore, in the final stage of the project, we need to connect the warning system with the VR headset, displaying written instructions on the screen to help users reach their desired points while avoiding singularities.

 

Tasks and Duties:

  1. Understanding the Kinematic Model (e.g., in MATLAB):

· Model illustrates the relationship between joint angles and the position/orientation of the end-effector.

  1. Understanding Forward Kinematic Analysis:

· Analyze to determine a singularity-free workspace.

  1. Implement Inverse Kinematics Solver:

· Develop a solver to compute the joint angles needed to reach a desired end-effector position/orientation, considering kinematic constraints and avoiding singularities. Instead of opting for a complex analytical solution for the 9 degrees of freedom (DOF) in 3D space, explore the utilization of a machine learning algorithm, such as ANFIS (Adaptive Neuro-Fuzzy Inference System), which combines Artificial Neural Networks (ANN) and Fuzzy Logic. This approach allows for solving the model with desired input/output coordinates/angles.

4. Singularity Detection:

· Identify singularities by examining the Jacobian matrix for ill-conditioned configurations.

  1. Singularity Avoidance:

· Devise strategies to adjust end-effector trajectory, reconfigure arm joint angles, or steer clear of workspace regions prone to singularities (comparing current end-effector position with predicted singularity points using robot live sensor data).

6. ROS Integration:

· Integrate the kinematic model, F/K kinematics solver, singularity detection, avoidance, and trajectory planning algorithms into ROS nodes, utilizing ROS topics and services for communication and movement control.

7. Develop a Simulation model for visualization:

· Employ simulation, e.g., in Gazebo, to test the algorithm.

  1.  Create a VR notification system:

· Connect ROS to the headset using a preferred coding framework

  1.  Controlling movement by Down/Upscaling

· Design and implement an adjustable mechanical control button for changing the scale factor (Possible software solution)

  1.  Verification:

· Conduct experimental tests to validate the proposed approach.

 

 Requirements:

o   Proficiency in MATLAB

o   Strong understanding of Kinematics/Robotics

o   Basic familiarity with ROS

o   Interest in machine learning algorithms

o   Interest in simulation and modeling

o   Interest in working with the Virtual Reality framework

 

Contact Person: Mohammad Sadeghi (mohammad.sadeghi@tuhh.de)

Desired starting date: As Soon As possible

Institut für Mechatronik im Maschinenbau (iMEK), Eißendorfer Straße 38, 21073 Hamburg