Will GPT-like AI technology reshape the technical route of autonomous driving?

This week, Alibaba’s “ChatGPT” Tongyi Qianwen was officially released, becoming the second large language model product similar to ChatGPT in China. At a time when artificial intelligence technology at home and abroad is exploding, large models like GPT are developing rapidly. Expand and extend into our lives.

At present, the smart cockpit part of many car models has announced the integration of artificial intelligence large models. In terms of intelligent driving, there is not much progress in the GPT category, but next week, Haomo Zhixing’s autonomous driving generative large model DriveGPT will be released. Driving will also officially enter the era of GPT blessing. Will the technical route of autonomous driving be reshaped?

DriveGPT can make autonomous driving come to fruition faster, instead of reshaping the technical route?

The DriveGPT autonomous driving cognitive large model also uses the RLHF (Human Feedback Enhancement Technology) algorithm to continuously optimize the autonomous driving decision-making model by continuously inputting real human driving data. In other words, it will draw on thousands of real driving data, thereby Make your own optimal solution. The overall training logic and algorithms used by DriveGPT are generally similar to ChatGPT, but the fields are different, but only under the current conditions, they are not the same.

With the development of artificial intelligence technology, the research and development and implementation process of autonomous driving have been accelerated. Autonomous driving needs to be realized through the three steps of "seeing, thinking, and acting". Among them, the "thinking" step is the most challenging, because it requires considering various complex situations and changes, such as traffic rules, road conditions, weather, etc. This requires building a powerful model to simulate the human thinking process, and GPT-like AI software was born for this purpose.

In autonomous driving training, GPT-like AI software can play a vital role. First, it can extract valuable information from massive data through technologies such as deep learning and natural language processing, classify, analyze and summarize this information, and then generate high-quality text and speech output. These outputs can be used to solve various problems in autonomous driving, such as identifying road signs, signs, and traffic lights. In addition, GPT-type AI software can also help self-driving vehicles predict the behavior of other vehicles and pedestrians and respond accordingly, such as emergency braking or steering.

In the field of autonomous driving, the amount of data generated, calculated and referenced is huge, requiring a lot of processing and analysis. GPT-like AI software has powerful computing and data processing capabilities, and can quickly process and analyze large amounts of data and generate useful conclusions and output. These conclusions and outputs can be used to develop better traffic rules and flexible driving strategies, thereby improving the efficiency and safety of the entire road system.

In addition to these most basic functions, GPT-like AI software can also be customized and developed according to user needs to adapt to different scenarios and needs. For example, when driving in a city, autonomous vehicles need to quickly identify obstacles ahead and take appropriate measures to avoid collisions. In this case, GPT-like AI software can strengthen the control and management of autonomous driving by predicting the location, speed and direction of obstacles. Although this type of predictive function has been used in autonomous driving, the data The scale is very limited. After accessing the GPT database, the redundancy will become higher.

At the same time, GPT-type AI software can also help autonomous vehicles learn and adapt to new road conditions and environments. For example, when a vehicle enters a new geographical area, it may encounter some new road signs and traffic signals and need to relearn how to process these signals and instructions. In this case, GPT-like AI software can provide valuable guidance and support, especially in areas without high-precision maps. The computing power of GPT, combined with the original perception and data processing, will make it possible for people in areas without maps to Autonomous driving becomes easier to achieve.

It is worth mentioning that GPT-like AI software can also help reduce human errors and unnecessary decisions. Self-driving vehicles need to be highly alert at all times and make correct decisions to avoid potential dangers. However, human error and misjudgment remain inevitable problems. By using GPT-like AI software, we can improve the accuracy and reliability of decision-making, thereby avoiding accidents and losses caused by human factors. AI software can also monitor the status and behavior of the autonomous driving system in real time, detect abnormalities and risks, and take appropriate measures in a timely manner.

AI itself doesn’t know where the boundaries are, so human intervention may not be useful?

In addition, GPT’s natural language processing technology can also be used to interact with passengers, understand their needs and preferences, and enhance safety and security measures during driving. Because when you want to communicate with ChatGPT, any of your questions can put the software into a specific context, so that it can answer questions more smoothly and naturally. After the GPT-type self-driving AI is familiar with the driver's habits, it can also It is equivalent to entering the "context", and its response method will be closer to the driver of the vehicle, but this also needs to give the AI ​​an evaluation standard. If the driver's driving habits are not good, then in this "context" , the performance of AI may be counterproductive.

Nowadays, with the continuous development of vehicle-road collaboration technology, more and more car models or apps are beginning to connect to the vehicle-road collaboration system. When artificial intelligence is added to the vehicle-road collaboration system, the vehicle's predictive perception capabilities will become stronger. Of course, this is possible It will also bring some uncertainty. When artificial intelligence escapes from the vehicle, or the backend computing center of the car company, and is connected to a wider amount of data or carrier, we don’t know what the AI ​​will do. However, we all understand that AI cannot be limited to a specific carrier or space. It is to connect everything in some way. As we become more convenient and technology development becomes faster, At the same time, beware of hidden dangers.

Everything has two sides. Just like other technologies, GPT-type AI software also has some potential risks and challenges in autonomous driving training. First of all, it requires a large amount of data for training and learning, so the quality and accuracy of the data have a crucial impact on its performance. You will also find in ChatGPT that it will make some obviously wrong answers. This requires Blame its database. It is very important for AI to know what is right, what is wrong, and where the bottom line is. Human beings also play a key role in this. Secondly, since self-driving vehicles must meet strict safety standards, GPT-type AI software needs to be fully tested and verified to ensure its safety and reliability. For example, Tesla currently uses AI software to train self-driving cars. software, so we will also see that Tesla’s autonomous driving standards may not adapt to the standards of various countries around the world.

Summarize:

Artificial intelligence plays an important role in autonomous driving training. It can help us establish accurate and reliable driving models, process and analyze large amounts of data quickly and efficiently, realize intelligent driving decisions, and establish multiple safety guarantee mechanisms. However, In fact, it is an even more powerful function. It only makes technology development faster in terms of some calibrations or detailed calculations. It is not a reshaping of the current technical route for autonomous driving research and development.

However, as we mentioned above, AI software such as DriveGPT is artificially controlled in the field of autonomous driving. If it is connected to more carriers, the amount of learning data can be expanded to more In terms of aspects, it may really reshape the technical route. Not only will it reshape, but it may even pose some threats to us. How to control AI artificial intelligence technology within a reasonable range will be what mankind will face together in the future. A big problem, a one-size-fits-all approach to "unplugging the network cable" and suspending development cannot stop a beast that is growing every moment.