APF+: Boosting adaptive-potential function reinforcement learning methods with a W-shaped network for high-dimensional games

2025-03-17

AI Summary

Adaptive Potential Functions (APFs) struggle to transfer effectively to high-dimensional pixel inputs—such as those in Atari games—due to the absence of ground-truth state supervision. Method: This paper proposes APF+W-Net, a novel framework featuring a W-shaped dual-branch encoder (W-Net) that jointly models visual state ontology and its representational bias. Integrated with contrastive representation learning (ST-DIM), Dueling DQN, and APF, it yields the APF-WNet-DDQN algorithm—enabling end-to-end pixel-level RL without access to true state labels. Contribution/Results: On 20 Atari games, APF-WNet-DDQN significantly outperforms baseline DDQN (winning in 14/20 games) and APF-STDIM-DDQN (13/20 wins), while matching the performance of APF-ARI-DDQN—which relies on privileged RAM-state supervision. This demonstrates that visual APF modeling achieves competitive, label-free representation learning with strong generalization and empirical efficacy.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Studies in reward shaping for reinforcement learning (RL) have flourished in recent years due to its ability to speed up training. Our previous work proposed an adaptive potential function (APF) and showed that APF can accelerate the Q-learning with a Multi-layer Perceptron algorithm in the low-dimensional domain. This paper proposes to extend APF with an encoder (APF+) for RL state representation, allowing applying APF to the pixel-based Atari games using a state-encoding method that projects high-dimensional game's pixel frames to low-dimensional embeddings. We approach by designing the state-representation encoder as a W-shaped network (W-Net), by using which we are able to encode both the background as well as the moving entities in the game frames. Specifically, the embeddings derived from the pre-trained W-Net consist of two latent vectors: One represents the input state, and the other represents the deviation of the input state's representation from itself. We then incorporate W-Net into APF to train a downstream Dueling Deep Q-Network (DDQN), obtain the APF-WNet-DDQN, and demonstrate its effectiveness in Atari game-playing tasks. To evaluate the APF+W-Net module in such high-dimensional tasks, we compare with two types of baseline methods: (i) the basic DDQN; and (ii) two encoder-replaced APF-DDQN methods where we replace W-Net by (a) an unsupervised state representation method called Spatiotemporal Deep Infomax (ST-DIM) and (b) a ground truth state representation provided by the Atari Annotated RAM Interface (ARI). The experiment results show that out of 20 Atari games, APF-WNet-DDQN outperforms DDQN (14/20 games) and APF-STDIM-DDQN (13/20 games) significantly. In comparison against the APF-ARI-DDQN which employs embeddings directly of the detailed game-internal state information, the APF-WNet-DDQN achieves a comparable performance.

Problem

• Extends adaptive potential function for high-dimensional games.
• Uses W-shaped network to encode game state representations.
• Improves reinforcement learning performance in Atari game tasks.

Innovation

• APF+ extends APF with a W-shaped network encoder
• W-Net encodes high-dimensional game frames into low-dimensional embeddings
• APF-WNet-DDQN outperforms baseline methods in Atari games