Beyond Liability: Decentralized Forensics for Autonomous Vehicle Events via a Redactable Blockchain Approach

Autonomous vehicles (AVs) can sense their environment and navigate without human input. However, when AVs are involved in accidents with other AVs or human subjects, liability must be indisputably determined based on accident forensics. However, existing methods rely on centralized authorities to collect and manage accident data, making them vulnerable to cybersecurity attacks (e.g., tampering, denial-of-service). Decentralized systems, such as blockchain networks, present a promising alternative; however, their data immutability properties present challenges in practice. Consequently, this paper introduces a redactable blockchain solution for vehicular event forensics. This novel approach, underpinned by a decentralized attribute-based chameleon hash function, supports modifying data from multiple blocks, as well as individual transaction-level modification operations, without the issue of a single point of failure. We implemented the proposed system using the Charm cryptography library and the Hyperledger Fabric blockchain platform. The Chameleon hash function can support redactable operations for on-chain vehicular event records with a latency of seconds. We also conducted extensive experiments on blockchain performance and the witness group formation cost, highlighting that the proposed redactable blockchain is efficient in practice for vehicular event forensics tasks.