An Adaptive Sharded Blockchain Architecture for Secure Autonomous Intelligent Systems via Collaborative Optimization

Abstract

Autonomous Intelligent Systems (AISs), such as smart vehicles, industrial robots, and healthcare monitors, are increasingly deployed in dynamic and distributed environments where secure, trustworthy, and decentralized coordination is essential. Blockchain technology holds significant promise for AISs by ensuring the integrity and traceability of sensing, control, and decision-making across multiple agents. However, conventional blockchain architectures suffer from limited throughput, and unpredictable latency, rendering them unsuitable for resource-constrained, latency-sensitive AIS scenarios. In such settings, timely data exchange and low-latency consensus are crucial to enabling real-time autonomy and coordination. To address these challenges, we propose COShard, a collaborative optimization architecture that co-designs sharded blockchain consensus and network resource allocation for AISs. COShard jointly adapts the shard count and allocates routing paths and bandwidth to maximize blockchain throughput under constrained communication resources, while guaranteeing bounded latency for critical AIS tasks. The optimization problem is iteratively solved by addressing two sub-problems: determining the optimal number of shards and allocating network resources efficiently. Furthermore, to handle the dynamic and complex nature of AIS networks, we use Graph Reinforcement Learning (GRL) for the resource allocation step. Simulation experiments demonstrate that our method significantly improves the throughput of the blockchain system and guarantees latency requirements compared to traditional methods.