Continuing what I learned about robots in a day which would have taken months - part 2
1. Proper Finetuning data is really important for training. Try to use leader follower arms for teleop data
2. Especially for hackathons, a task can be done in a lot of ways, so try to think out of the box and solve the problem statement without any disqualifications
3. Using Astra like models for harness doesn’t work in realtime because it might take around 10 sec to get result from ur harness and till then the environment might be changed
4. Try to use conventional computer vision models or jev image models for object recognition (if u don’t wanna finetune yolo nano models)
5. Major issue with robotic cameras is the original colors of objects in the video and they would not be the same which you see using your naked eye. The yellow u see might be green in the video due to cameras
6. Last but REALLY REALLY important thing to check if all your cameras are stable or else all the data collected will be wrong.
1. Proper Finetuning data is really important for training. Try to use leader follower arms for teleop data
2. Especially for hackathons, a task can be done in a lot of ways, so try to think out of the box and solve the problem statement without any disqualifications
3. Using Astra like models for harness doesn’t work in realtime because it might take around 10 sec to get result from ur harness and till then the environment might be changed
4. Try to use conventional computer vision models or jev image models for object recognition (if u don’t wanna finetune yolo nano models)
5. Major issue with robotic cameras is the original colors of objects in the video and they would not be the same which you see using your naked eye. The yellow u see might be green in the video due to cameras
6. Last but REALLY REALLY important thing to check if all your cameras are stable or else all the data collected will be wrong.
Today I learned a lot about robotics which I might have not learned in months
1. Doing teleportation for robotic arms is really difficult using meta quest if you don’t have experience playing games😅
2. Force cannot be exactly designed in simulators and so training the grippers is kind of hard for objects which are not rigid bodies
3. Tool center point really affects your robotic operation
4. Specialized Harness helps a lot if you are using @dreamlabs_ai
5. Sometimes your robot will just give up in you because of servos heating or maybe the hardware has slight electrical communication issues
6. @dreamlabs_ai has some really good apis which you can use for different robotic arms and @makermodsai has the robots for you to try them and learn more within hours not months.
7. Robots might break things if you are not careful 😂
Many more insights learnt..View quoted post on X