I have a 3d nifti image i want to path plan in it using rrt ,it has only labels 0,1,2,3 where 1,2,3 are obstacles, how do i define statspace for this?

The image is segmentation label map how do i convert into a map that robot can traverse?

Answer 1

[Balavignesh]

Hi Mukund,

It is my understanding that you would like to perform path planning using Rapidly-exploring Random Trees (RRT) for a 3D 'NIfTI' image. You need to first convert the image into a format that defines the free space and obstacles in a way that the RRT algorithm can understand and operate on. Assuming your image has labels where 0 represents free space and 1,2,3 respresent obstacles, you could create a binary occupancy map where 0s are free space and 1s are obstacles.

For RRT in a 3D environment, your state space will consist of positions in 3D space that the algorithm can explore. Since MATLAB's RRT functions (if you're using Robotics System Toolbox) do not directly support 3D occupancy grids, you might need to implement or adapt RRT for a 3D grid environment yourself.

Implementing RRT from scratch or adapting an existing 2D version to 3D involves:

• Randomly sampling points in the 3D space defined by your state space.
• Extending the tree from the nearest node towards the sampled point, within a certain step size, ensuring the path does not collide with obstacles.
• Repeating this process until you reach the goal region or exceed a maximum number of iterations.

Here is a very basic skeleton of what an RRT algorithm might look like.

function path = rrt3D(obstacleMap, startPose, goalPose)
    % Initialize RRT tree with start pose
    tree = startPose;
    
    % Define maximum iterations
    maxIterations = 1000;
    
    for i = 1:maxIterations
        % Sample random point in 3D space
        randomPoint = [randi(size(obstacleMap, 1)), randi(size(obstacleMap, 2)), randi(size(obstacleMap, 3))];
        
        % Modify all the below functions as per your implementation 
        
        % 1) Find nearest node in the tree to the random point
        % NearestNode = findNearestNode(tree, randomPoint);
        
        % 2) Attempt to add a new node in the direction of the random point
        % newNode = extendTree(NearestNode, randomPoint, obstacleMap);
        
        % 3) Check if goal is reached or close enough
        % if isGoalReached(newNode, goalPose)
        %     break;
        % end
    end
    
    % Construct path from start to goal
    % path = constructPath(tree, startPose, goalPose);
end

Once you have your RRT implementation ready, you can integrate it with the rest of your application, feeding the start and goal poses into the algorithm and using the resulting path for navigation or analysis.

Kindly have a look at the following documentation links to have better understanding on:

• 'geometry planning using RRT' : https://www.mathworks.com/help/nav/ref/plannerrrt.html
• 'randi' funtion : https://www.mathworks.com/help/parallel-computing/codistributed.randi.html

Hope that helps!

Balavignesh