Hardware & Making
A Competitive Time-Trial AI for Need for Speed: Most Wanted Using Deep Reinforcement Learning
All challenges and achievements in creating a competitive time-trial AI in NFS:MW.
15 years ago, at the height of my eSports career, I uploaded an (unofficial) ESL record at Need for Speed: Most Wanted (2005) (NFS:MW) to Youtube. In the meantime Deep Reinforcement Learning became popular and ever since I have dreamt of creating a competitive AI for my favorite racing game of all time: NFS:MW. Now finally the time was right: The hardware is fast enough, good software is available, and Sony's AI research has proven the task is actually doable. Hence I thought: "How hard can it possibly be?".
This talk will present in detail all challenges and achievements in creating a competitive time-trial AI in NFS:MW from scratch - including but not limited to - hacking of the game to create a custom API, building a custom (real-time) OpenAI gym environment, steering the game using a virtual controller, and finally successfully training an AI using the Soft-Actor-Critic algorithm. All code including the API is written in Python and is open source.
Additional information
| Live Stream | https://streaming.media.ccc.de/38c3/zigzag |
|---|---|
| Type | Talk |
| Language | English |
More sessions
| 12/27/24 |
I'm not big-brained enough to use cameras on Linux, so I decided to write my own camera stack (based on a real story).
|
| 12/27/24 |
Reverse engineering the Wi-Fi peripheral of the ESP32 to build an open source Wi-Fi stack.
|
| 12/27/24 |
Many developers know that the answer to "How do I debug this microcontroller" is either "JTAG" or "SWD". But what does that mean, exactly? How do you get from "Wiggling wires" to "Programming a chip" and "Halting on breakpoints"? This talk will cover how common debug protocols work starting from signals on physical wires, cover common mechanisms for managing embedded processors, and ending up at talking to various common microcontrollers.
|
| 12/27/24 |
The Iridium satellite (phone) network is evolving and so is our understanding of it. Hardware and software tools have improved massively since our last update at 32C3. New services have been discovered and analyzed. Let's dive into the technical details of having a lot of fun with listening to satellites.
|
| 12/27/24 |
The 530 tons and 63 meter tall Ariane 6 rocket finally launched on July 9th 2024 carrying our open-source developed payloads – the SIDLOC experiment and the satellite Curium One – into space. SIDLOC tested a new, open, low-power standard for identifying and precisely locating spacecraft whilst our satellite Curium One established an open-source baseline for larger CubeSat systems and allowed us to test a bunch of new technologies. From sourcing a launch opportunity to the final integration ...
|
| 12/27/24 |
Recent breakthroughs in machine learning have dramatically heightened the demand for cutting-edge computing chips, driving advancements in semiconductor technologies. At the forefront of this progress is Extreme Ultraviolet (EUV) lithography—a transformative method in microchip fabrication that enables the creation of ultra-small, high-performance devices. However, the path from raw materials to these state-of-the-art chips navigates a complex global supply chain riddled with technical ...
|
| 12/27/24 |
Custom silicon chips are black boxes that hold many secrets, like internal ROMs, security features and audio DSP algorithms. How does one start reverse engineer them? Let's look at the basics of silicon reverse engineering, what gate array chips are, and how some tooling can generate Verilog code automatically from a die shot.
|
