What it does
The goal is that there are 4 checkpoints along the corridor and the robot with the Comma 3 device should be able to navigate through each checkpoint autonomously
How we built it
We used a Mobile SAM to segment the carpet along the corridor and this segmented information was fed to a Simple NN. which decides whether to go forward, left or right.
Challenges we ran into
The main issue we ran into was with datasets. During the collection of a dataset based on the commands we give on the manual drive, some images were captured in the wrong labeled folder due to latency issues which made our model not not reach good accuracy. we had to iterate on each dataset to reclassify the datasets for the training.
What we learned
- data curation is pivotal
- Choosing the right model is very important
- Make sure the latency of the image streamed from the device is minimal
What's next for Senor Todd's navigation
Navigation based on Mapping
- Here the idea is to use the device to drive around and create a map.
- Use the created Map and Nav2 to create a waypoint based navigation to each checkpoints.
- Add obstacle avoidance to be included in the local planner by using Yolo to detect and avoid people.
Log in or sign up for Devpost to join the conversation.