lexfridman/deeptraffic

DeepTraffic is a deep reinforcement learning competition, part of the MIT Deep Learning series.

deeptraffic's Issues

Most of the links provided do not work

Most of the links provided in the readme file do not work. Are there any other link we can access the material?

submission size limit

For some time now, the website https://selfdrivingcars.mit.edu/deeptraffic/
is coming up with a very low submission size limit,
for example:
Submission size: 672689 (model + code). Limit: 100000.
and refuses to accept any of my previous submissions or similar variants.

Is this intentional?

Error on the leaderboard page

I noticed that the leaderboard page cannot display the rank list. Perhaps there is some problem with the PHP code according to the information.

Basic Code

Thanks for the course. I am currently following it. I am new to AI. I want to take part in the competition. Is there some basic running code that I can take and start working on it? Sorry if it is not a relevant question.

Compatible to OpenAI Gym?

Namaste!

Great work! I really like it!

Question: Is DeepTraffic compatible to OpenAI Gym? I remember Nvidia writing about it in a 2017 blog-article.
https://blogs.nvidia.com/blog/2017/07/07/deeptraffic-how-an-mit-simulation-game-uses-deep-learning-to-reduce-gridlock/

Best,
Tristan

Observation error?

Hi!

From simple investigation of observations passed to the neural network's input, I see very weird values passed having the following issues:

First of all, observation vector contains 0 for cells outside of the road and 1 if cell is not occupyed by other car (those values are not mentioned in arxiv paper).

Second, much more serious problem is related to situations with occupied cells. Ff the cell is occupied by other car, it doesn't have speed relative to the central car (as was stated in the paper), but rather has values between 0.018 and 0.27, regardless of relative speed of the other car.

Even if my car is constantly issuing decelerate action and all the cars are overtaking me at high speed, this number is positive and has the range above. For example, that's the small log of state vector I'm observing if lanesSide = 3, patchesAhead = 1, patchesBehind = 0, temporal_window = 0 (which gives me just 7 numbers for row in front of the car):

VM1698:71 (7) [1, 0.025997055518438162, 1, 1, 1, 1, 1]
VM1698:71 (7) [1, 1, 1, 1, 0.020270829044069006, 1, 1]
VM1698:71 (7) [1, 0.02407384558168598, 1, 1, 1, 1, 1]
VM1698:71 (7) [1, 1, 1, 1, 0.026940491870455902, 1, 1]
VM1698:71 (7) [1, 1, 1, 1, 1, 1, 1]
VM1698:71 (7) [1, 1, 1, 1, 1, 1, 0.02142199183294735]
VM1698:71 (7) [1, 1, 0.018471935100245227, 1, 1, 1, 1]

As I've said, those numbers stay positive regardless of relative speed, which looks like a bug and deviates from formula given in the paper.

Milehigh.world

At Milehigh.world, we’re pioneering a revolutionary software application that harnesses the power of deep learning to redefine music exploration in the tech world.

Picture a cutting-edge tool that transcends traditional boundaries, delving into the infinite realms of music with unparalleled precision and insight. Our software leverages state-of-the-art deep learning algorithms to unravel the complexities of musical composition, opening doors to endless creative possibilities.

With MileHigh.world, users immerse themselves in a world where technology meets innovation, where data-driven insights fuel artistic expression, and where the boundaries of music are continually pushed and expanded. Our software analyzes vast datasets of music, identifying patterns, trends, and hidden connections that lay the foundation for groundbreaking discoveries.

But MileHigh.world isn’t just about analysis—it’s about empowerment. Our software equips users with intuitive interfaces and powerful tools that enable them to become architects of sound, shaping and sculpting melodies with unprecedented control and precision. Whether you’re a seasoned programmer or a curious enthusiast, our software offers a playground for experimentation and discovery.

What sets MileHigh.world apart is our relentless pursuit of innovation and collaboration. We’re not just building a software application; we’re creating a community of like-minded individuals who share a passion for pushing the boundaries of technology and music. Through collaborative projects, shared insights, and ongoing research, we’re forging new pathways and redefining what’s possible in the world of music and technology.

Join us on a journey to the forefront of music exploration in the tech world. Together, let’s unlock the infinite possibilities of deep learning and pave the way for a future where creativity knows no bounds.

DeepTraffic not available

DeepTraffic is no longer available at https://deeplearning.mit.edu/deeptraffic . Is this permanent? Temporary with a new version soon? If a new version is coming, when?

Thank you in advance for any information you can provide.