The Bidding Games: Reinforcement Learning for MEV Extraction on Polygon Blockchain

TL;DR

The paper tackles high-frequency, sealed-bid MEV auctions on Polygon Atlas by introducing a PPO-based reinforcement-learning bidding agent trained in a realistic simulator that models stochastic opportunity arrivals and competition. It demonstrates that a history-conditioned agent can capture substantial profits, achieving up to 49% of available profits against existing searchers and up to 81% when replacing the market leader, outperforming static bidding strategies. The work provides practical implications for industrial participants and protocol designers by showing that adaptive, temporally-aware bidding strategies are essential under sub-second, partially observable conditions. It also outlines deployment considerations and future extensions to other MEV strategies and cross-chain opportunities.

Abstract

In blockchain networks, the strategic ordering of transactions within blocks has emerged as a significant source of profit extraction, known as Maximal Extractable Value (MEV). The transition from spam-based Priority Gas Auctions to structured auction mechanisms like Polygon Atlas has transformed MEV extraction from public bidding wars into sealed-bid competitions under extreme time constraints. While this shift reduces network congestion, it introduces complex strategic challenges where searchers must make optimal bidding decisions within a sub-second window without knowledge of competitor behavior or presence. Traditional game-theoretic approaches struggle in this high-frequency, partially observable environment due to their reliance on complete information and static equilibrium assumptions. We present a reinforcement learning framework for MEV extraction on Polygon Atlas and make three contributions: (1) A novel simulation environment that accurately models the stochastic arrival of arbitrage opportunities and probabilistic competition in Atlas auctions; (2) A PPO-based bidding agent optimized for real-time constraints, capable of adaptive strategy formulation in continuous action spaces while maintaining production-ready inference speeds; (3) Empirical validation demonstrating our history-conditioned agent captures 49\% of available profits when deployed alongside existing searchers and 81\% when replacing the market leader, significantly outperforming static bidding strategies. Our work establishes that reinforcement learning provides a critical advantage in high-frequency MEV environments where traditional optimization methods fail, offering immediate value for industrial participants and protocol designers alike.

Figures (6)

• Figure 1: Empirical transaction execution flow observed in the Polygon Atlas network.
• Figure 2: End-to-end architecture of a professional Polygon MEV searcher system, illustrating data ingestion, opportunity detection, strategy formulation, execution pipeline, and sealed-bid submission mechanics.
• Figure 3: Weekly statistics on unique MEV searchers, showing consistent participation from 5-8 core entities with occasional peripheral activity.
• Figure 4: Temporal activity patterns of a professional searcher's addresses, showing parallel bot deployment, operational timeline, and bid escalation patterns. Address annotations indicate mean bribe percentage and participation count.
• Figure 5: Bribe percentage distribution by protocol type, showing inverse relationship between opportunity frequency and mean bid amount. Complex protocols attract fewer but more aggressive bidders.