How do you develop AI and robotics projects?

I had a similar question, but it was along the lines of 'How do I add AI to my robotics project?' After some research, I realized my approach was misguided because my understanding of AI was incorrect. I needed to clarify my problem and approach, considering what I truly want from this project. Some possible approaches include: 1. Connecting a computer running the AI to a robot powered by a microcontroller via serial communication. 2. Having a single computer (like a Raspberry Pi) run both the robot and the AI. 3. Training a machine learning model and condensing it into a TinyML model that runs directly on a microcontroller. For a great example of what I had in mind, I'd suggest looking into IMO Desktop Robot It clarified my doubts.

When choosing a problem, it’s important to have a clear idea of the desire behavior. I often find that writing down the goal and trying to explaining the problem space to someone not related to it is quite useful. Find as many possible sources of information and data as possible, later in you’ll probably narrow down the list to only a few but at the beginning it’s best to have more options.

If you want to solve a problem with AI and robotics, you need to be smart and creative. You need to find a problem that is hard, but not too hard. A problem that is fun, but not silly. A problem that is cool, but not weird. Here are some ideas: - Think about what you like. What are some things that make you happy or curious? How can you make them better or more fun with AI and robotics? - Think about what you don't like. What are some things that annoy you or bore you? Maybe you don't like doing laundry, or driving, or cooking. How can you make them easier or faster with AI and robotics? - Think about what the world needs. What are some things that are important or urgent? How can you help or fix them with AI and robotics?

Framework selection for both technologies is key to ensure smooth integration after problem identification and project planning. Modular testing is also encouraged to overcome tedious troubleshooting as this is a complex undertaking

Indeed, having a clear strategy for implementing AI in Robotics projects is crucial. For instance, when I integrated AI into my humanoid robot ROBA for tasks like pick-and-place operations, I carefully considered various Machine Learning (ML) and Deep Reinforcement Learning (DRL) processes. This involved real-time object detection and depth image processing to precisely locate objects for manipulation. Additionally, DRL was employed for tasks such as inverse kinematics and motion planning for the robot's hands/arms.

After defining the problem, I do a literature review to scout current solutions on online platforms like GitHub, Google Scholar, or simply Google. The aim is to filter a vast variety of solutions to a handful that you can adopt, apply, and improve.

Use tools to find similar papers related to the problem space you are interested on. Arxiv.org is your friend, understand how others approach a similar problem (or maybe the same) and use that to build your framework of thought. Check what others have already tried and see what does work and what does not. Paperswithcode.com is another good resource to check!

Before finalizing the frameworks, do a quick analysis of what you are actually building. What hardware will be required in this robot, for it to successfully execute your desired task. For example, your input device, is it a camera? an ultrasonic sensor? or a combination of both and so on, based on that one can do quick search for existing frameworks to give an overall idea. Finalizing a SBC like RaspberryPi or Jetson boards will be post that according to your projects demands. Post that the toolset can be decided on similar lines.

Statistically, 67% of tech firms prioritize AI ethics (IEEE, 2024). In my opinion, modular prototyping is key. One thing I've learned is that using tools like Gazebo can streamline development. I've found helpful that combining ethical AI frameworks ensures inclusivity. Common mistakes include neglecting bias audits. I disagree with ignoring community input; diverse perspectives enhance outcomes. The future lies in ethical, innovative AI systems designed for social justice.