350 points by ai_engineer 6 months ago flag hide 12 comments
user1 6 months ago next
How did you ensure the safety of your RC car during training and testing? I'm worried about collisions.
author3 6 months ago next
Safety was a top priority for me. I used a combination of sensors, including LIDAR and cameras, to detect obstacles in the car's path. I also used a simulated environment for initial testing to minimize the risk of physical damage.
author1 6 months ago prev next
Great job! I'm curious how you trained the deep learning model. What dataset did you use?
author2 6 months ago next
I used a custom dataset that I collected using a combination of open-source data and data that I generated myself. I annotated the data using a professional data labeling service.
user2 6 months ago prev next
What kind of hardware are you running the deep learning model on? How did you optimize the model to run in real-time?
author4 6 months ago next
I'm using a NVIDIA Jetson Xavier NX for the hardware. I optimized the model using tensorRT to run efficiently on the device. I also used a small convolutional neural network to keep the computational overhead to a minimum.
user3 6 months ago prev next
How did you handle the problem of the car getting stuck in local minima or making suboptimal decisions?
author5 6 months ago next
I used a combination of reinforcement learning and imitation learning to train the model. The reinforcement learning component helped the model to escape local minima, and the imitation learning component helped the model to make better decisions by learning from human drivers.
user4 6 months ago prev next
Have you considered open-sourcing the code and data? I'm sure many people in the community would be interested in replicating your work.
author6 6 months ago next
I have thought about open-sourcing the code, but I'm not sure if I want to make the data publicly available due to copyright issues. I might consider sharing the code with a few researchers who are interested in collaborating.
user5 6 months ago prev next
What kind of performance improvements did you see by using deep learning instead of traditional control algorithms?
author7 6 months ago next
Deep learning allowed me to train the model to handle a wide variety of driving scenarios, some of which were difficult to model using traditional control algorithms. I also saw improvements in the car's ability to navigate around obstacles and avoid collisions.