The aim of my project is to test the efficiency, advantages and disadvantages of 3-4 algorithms selected in two environments by using them on an autonomous delivery robot. These environments;
The first environment is outdoor, dynamic (there will be objects to move other than the robot) and online. The second environment will be an indoor and dynamic environment. (So the only difference is that they are outdoor and indoor)
Path planning algorithms are required to compare the autonomous mobile delivery robot in two different conditions. I have searched the literature and reviewed numerous articles, but the more I read, the more information emerges. Finally, I decided to choose and compare one of the algorithms under the headings of Deterministic and Probabilistic Path Planning, Grid-Based and Measurement-Based Road Planning, Mathematical and Intuitive Path Planning (I would like to hear your ideas). However, I encounter a lot of information intertwined with each other.
Are there any path planning algorithms you can recommend me to use? I would be interested in the resources and articles you can recommend.
I hope you can suggest path planning algorithms that I can use to compare in my project.
Use " RRT" or " RRt* ", however it is still fuzzy to answer untill knowing the excat scenario ... good luck
Mirza Türkyilmaz
This sounds like an excellent robotics project comparing path planning algorithms. I'd be happy to suggest some well-suited approaches to consider:
- For the outdoor environment, I would recommend A* grid search for deterministic completeness and simplicity. Can incorporate a heuristic like distance to goal for efficiency.
- For indoor, a rapidly-exploring random tree (RRT) algorithm leveraging obstacle proximity and random sampling could be effective for complex spaces.
- Model-based methods like vector field histogram (VFH) or dynamic window approach (DWA) offer intuitive performance based on velocity obstacles.
- Machine learning approaches like reinforcement learning or neural networks to train the robot to navigate autonomously through diverse examples.
I would focus on 1-2 algorithms from the grid/graph search space and 1-2 from the measurement-based category to showcase the continuum of planning philosophies. The key will be robust obstacle detection and mapping integration.
Please feel free to reach out if you need any implementation tips or additional literature sources. It's an exciting project direction with high relevance. Wishing you the very best.
Machine learning approaches such as reinforcement learning or neural networks for robot training.
Hello Mirza Türkyilmaz . In our laboratory we make use of a planning algorithm based on the eikonal equation of light. The algorithm is called Fast Marching Square and we have applied it to planning in 2D and 3D Cartesian environments for mobile robots, drones and manipulation (where we have developed a learning-by-imitation algorithm based on this idea). It is a good technique, it allows to take objects in the environment into account and it has several advantages over other planning algorithms such as the absence of linear minima or that it is always able to find a solution (it is a complete algorithm). The only problem is the "curse of dimensionality". In more than 3 dimensions the algorithm becomes extremely slow and even in high dimensions (e.g. if you want to plan in the configuration space of a robotic arm) it is unable to generate a solution as it is not able to converge. If you have any questions, please do not hesitate to contact me.
- Badges
- Science topic