“Autumn Mountain is sparsely pedestrian, and there are often higher and lower drivers. Nowadays, unmanned vehicles are on the road, and they are all AI veteran drivers.”
Let me ask how the old AI drivers perform?
It can turn flexibly and control speed freely:
Feasible clouds and flowing water, turning along the optimal route:
You can also make multiple loops, and it doesn't matter how many times you go around:
In this racing event, the protagonist is indeed a group of veteran AI drivers. Humans can only watch the whole process, and there is no remote control to participate.
More precisely, they are artificial intelligence models trained through tens of millions of virtual training sessions. The competitions these models compete in are also quite prestigious in the industry. This is what Amazon Cloud Technology holds——
Amazon DeepRacer League。
As the autopilot racing league with the most developers in the world, Amazon DeepRace does not fight who has the money to buy a sports car, and who does not have the resources to build a venue, only fights the reinforcement learning algorithm and training level written by the players——
It can be called a variety show of global intelligence level, fully interpreting the true meaning of "Smart is the new sexy".
Since its inception, Amazon DeepRacer has been quite well-known in the AI/ML circle, with 140,000 developers participating, and Zhihu also has iron fan experience and Po certificates.
The influence of the public has increased, and only one trailer has 3.8 million views.
As the race gets more and more laps, last year, Amazon DeepRacer League and F1 official co-branded, held a virtual "Barcelona" race, to become the F1 leader in the autopilot industry.
So, what is the origin of this Amazon DeepRacer? Why make technology enthusiasts so top?
Autonomous driving league with the most developers
In 2018, Amazon Cloud Technology was born with the idea of holding an AI algorithm competition to let more people learn about related technologies and even become practitioners. The Amazon DeepRacer League came into being.
At the beginning of the event, it was aimed at non-industry groups. Almost all participating software resources were officially provided, and there were even tutorial guides. Participants did not even need to spend money to buy physical cars. They could first train their own autonomous driving algorithms on the cloud platform with a 3D racing simulator. .
After getting started, participants can sign up to participate in the monthly online qualifiers. If you rush to the top 64 in the world, you can qualify for the offline league——
Compete against your model with masters from all over the world.
Although the threshold is low, the prize money is also high.
The champion will receive a prize of US$10,000, the runner-up will receive a prize of US$5,000, and the third place will receive a prize of US$2,000.
With low threshold and high bonus, the game is still playing with cutting-edge technology. The competition has been held since it has attracted the attention of people from different backgrounds around the world.
Participants included Silicon Valley full-stack developers and pilots, employees from Singapore and Canadian banks, and Chinese university students...
The participation of participants from different backgrounds not only reflects the diversity of the game, and makes machine learning technology known to more people, but also adds many interesting fragments to the game.
For example, in the Amazon DeepRacer TV 2019 finals, a Taiwanese brother won the championship after reversing the victory, but when he gave his acceptance speech, he expressed his feelings in Chinese, so the English host didn’t know how to answer the call...
It is worth mentioning that this year, DeepRacer also began to host the China Challenge for Chinese developers in China. The prizes are also generous and worthy of attention and expectation.
How to tame an unmanned car?
Looking back at the technical theory from the game, in fact, the foundation behind the Amazon DeepRacer activity is simple and cutting-edge.
How to understand this sentence?
Might as well start with a self-driving car structure.
Amazon DeepRacer's self-driving cars are differentiated by generations. The first generation of cars is the main force in the competition. The core includes an Intel Atom® processor, a 4-megapixel 1080p resolution camera, a high-speed WiFi and an Intel OpenVino computer vision tool suite. The first generation of cars is mainly used in racing competitions, and the race track is judged through visual recognition to complete the race.
The higher-end is the second-generation DeepRacer Evo, equipped with a stereo camera and lidar, which can be used for chasing and obstacle avoidance competitions.
If you are familiar with autonomous driving, you may also be able to see that this configuration can be described as "simple", but it is enough to experience the fun of autonomous driving——
This is because it is the algorithm and model that truly give the car the soul of autonomous driving.
Like the self-driving cars on the road today, Amazon DeepRacer also uses a technology in the machine learning category:
Reinforcement Learning (RL).
This technology has only been in existence for a few years, but it can be regarded as one of the most important technological breakthroughs of mankind in recent years. Behind AlphaGo, which defeated Lee Sedol in 2016, this technology played a role.
The principle of reinforcement learning is not difficult to understand. It is a kind of "trainer" mechanism, and it can also be understood as the principle of "Pavlov's dog".
Take "chicken pecking paper" as an example:
If we want to train a chicken to peck a piece of pink paper, then every time it pecks a piece of pink paper, it will be rewarded with food.
After some training, the chicken has been able to find the pink piece of paper correctly.
from Twitter user @YAWScience
Of course, rewarding unmanned cars and computers can't use food, but use the "reward function" they can understand (Reward Function).
For Amazon DeepRacer, developers can set a super simple reward function, which has only two parameters: all\_wheels\_on_track indicates whether all four wheels are on the track, and progress represents the current completion progress.
Specific to the code, if a wheel does not happen once on the track, the reward function will be deducted, otherwise, it will always be rewarded with the progress of the car.
The larger the value of the reward function, the more successful the training. When the reward function is negative, the task fails.
def reward_function(params):
if not params["all_wheels_on_track"]:
reward = -1
else:
reward = params["progress"]
return reward
If you want the car to be smarter and don't turn around on the runway, we can also set the function to be a little more complicated, taking the distance of the car away from the dashed line of the center of the runway as a parameter. The smaller the distance, the greater the reward.
During the training process, the car records its position through a camera, etc., and continuously adjusts the speed and driving direction. When the reward function is maximized and stabilized——
"Beast Taming" is complete. Congratulations, you have mastered the basic principles of reinforcement learning.
How to enter?
The game is exciting and fun, and there are bonuses available. The principle is easy to understand. Now, can't you wait to play?
The way to participate is very simple: just register for an Amazon Cloud Technology account.
Log in to the account to enter the Amazon DeepRacer console.
Before constructing the model, configure a virtual race car of your own in the garage.
Give it a cool name (the name can't be changed, don't just choose "test" and it's over), choose a color you like and configure the camera according to the type of competition you want to participate in.
After creating the exclusive virtual car, follow the prompts to create the model step by step:
The official also provides a virtual environment to lower the threshold for getting started, and let the car train in the virtual world. After completion, you can submit the model to choose a competition to participate in!
Participating in online competitions does not require a physical car, just participate directly.
After you emerge in the field of virtual training, you can also get a free physical car from Amazon Cloud Technology, worth $300.
Of course, if you don’t need money, you can also buy it directly on the official website:
Although it costs a little money to buy a car, it does add to the fun. You can recreate the track at home and practice at any time.
It’s said that some people just changed to a big house in order to get a good result and reproduced the track practice in their underground garage (manual dog head)——
The official website also intimately provides a tutorial for building a track at home:
Finally, focuses on to provide some opportunities to gather official wool.
If you are a new user for the first time to participate in Amazon DeepRacer project, the first month you can get long 10 hours of free training and 5GB of storage space .
In addition to free time, Amazon DeepRacer also provides free video tutorial teach you the basics of reinforcement learning and guide to the competition.
Competition skills are also the focus. The official webpage even contains the strategies written by the lap record holders:
Therefore, or an experienced machine learning practitioner, , 161bbef335b840, you can feel confident and bold to participate.
Racing games let more people get started with AI
As mentioned earlier, Amazon Cloud Technology wanted to promote AI algorithms in game activities in 2018, and later Amazon Deepracer. The purpose of this move is to face the turbulent development of AI/ML. This big company hopes to let more people know about it. Related technologies, and then participate in it.
Take Amazon Deepracer as an example. In order to expand the number of participants, Amazon Cloud Technology also launched a student competition for high school and college students over 16 years old. Registration does not require a credit card to participate.
In addition, if you complete all the learning modules, get at least 80% of the scores in all the course tests, and have excellent single lap results in the student league, you can also get a scholarship. The top 500 students will also receive technical expert guidance from Amazon Cloud Technology and Intel for a period of 12 months.
Not only Amazon DeepRacer, but Amazon Cloud Technology allows people of different backgrounds to understand and get started with machine learning, and has actions in all aspects.
For example, the Amazon Cloud Technology AI&ML Scholarship Program that provides 10 million US dollars;
For example, the release of Amazon SageMaker Canvas, a graphical interface for machine learning without code, helps enterprise users with zero machine learning experience solve business problems, so that people of all business backgrounds can get started.
For another example, Amazon Sagemaker Studio Lab, a machine learning "lab" that provides free computing power for more developers...
I believe that only when more and more people understand and get started with AI/ML-related technologies and products, this wave that has only risen for less than 10 years can truly release its value.
For more information about the new release of Amazon Cloud Technology, take a long time to understand the picture below:
the end, do you want to try Amazon DeepRacer?
Add an official assistant for more details, the portal is here:
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用。你还可以使用@来通知其他用户。