#PPO
3 related articles
·3 min
Open-Source Robustness Testing Framework SIRFEX: Why Does Failure-Driven PPO Retraining Hurt Nominal Performance?
Open-source framework SIRFEX fine-tunes PPO on failure scenarios, yet nominal performance degrades. We analyze the experimental data, causes of negative results, and the balance between robustness and nominal performance.
Read more →
·2 min
PPO Algorithm Cracks the Moving Sofa Problem, Reconstructing Gerver's Optimal Solution
Developer uses PPO reinforcement learning to reconstruct Gerver's optimal solution to the Moving Sofa Problem, showcasing AI's potential in geometric optimization and its limitations.
Read more →
·3 min
Training a PPO Agent to Play Temple Run on a Phone: Real-Device Record Hits 16,782 Meters
A developer trained a PPO reinforcement learning agent to play Temple Run, using Frida to read game memory and call game methods directly. A custom C++ headless simulator trains 10M steps in 6 minutes. In 14 real-device tests, the median was 4,736 meters and the best was 16,782 meters. Now open-sourced on GitHub.
Read more →