IROS2019國(guó)際學(xué)術(shù)會(huì)議論文集 2303

Black Block Recorder Immutable Black Box Logging for Robots via Blockchain Ruffi n White1 Gianluca Caiazza2 Agostino Cortesi2 Young Im Cho3 and Henrik I Christensen1 Abstract Event data recording is crucial in robotics research providing prolonged insights into a robot s situational under standing progression of behavioral state and resulting outcomes Such recordings are invaluable when debugging complex robotic applications or profi ling experiments ex post facto As robotic developments mature into production both the roles and require ments of event logging will broaden to include serving as evidence for auditors and regulators investigating accidents or fraud Given the growing number of high profi le public incidents involving self driving automotives resulting in fatality and regulatory policy making it is paramount that the integrity authenticity and non repudiation of such event logs are maintained to ensure account ability Being mobile cyber physical systems robots present new threats and vulnerabilities beyond traditional IT unsupervised physical system access or postmortem collusion between robot and OEM could result in the truncation or alteration of prior records In this work we address immutablization of log records via integrity proofs and distributed ledgers with special considerations for mobile and public service robot deployments Index Terms Robot Safety Networked Robots Software Mid dleware Cryptobotics Distributed Ledgers I INTRODUCTION R OBOTS being cyber physical systems CPS are increas ingly deployed as part of a cyber infrastructure becoming ever more interconnected with the Internet of Things IoT The use of robots in a connected eco system is far from trivial How do we make these robots safe How can we verify correct operation How can we document operations for the purpose of traceability or document operations in the presence of failures An essential part of the design deployment and verifi cation of distributed robot systems is the ability to monitor and record runtime event data The possibility of deploying real world honeypots is a serious concern given the widespread interest in exploiting self driving cars and autonomous drones 1 Considering the recent history in automotive exploitation 2 there can be no doubt that lack of security represents a real threat 3 6 Documenting the operation of an integrated robotic sys tem composed of multiple components while generating a comprehensive trace of the operation is essential to quality control debugging systems verifi cation etc For debugging information fl ow or in cases of unexpected robot behaviour event logging is fundamentally integral However when the Manuscript received February 24 2019 Revised May 24 2019 Accepted July 8 2019 This paper was recommended for publication by Editor Paolo Rocco upon evaluation of the Associate Editor and Reviewers comments 1 Ruffi n and Henrik are with UC San Diego USA rwhitema ucsd edu 2Gianluca and Agostino are with Ca Foscari University of Venice Italy 3Young is with Gachon University Seoul South Korea Robot Enclave Planner Node LIDAR Node Log Storage Recorder Node Topic Data Checkpoint Stamp 1 Proof Checkpoint Stamp 1 Proof Checkpoint Stamp t Proof Checkpoint Stamp 1 Proof Checkpoint Stamp 1 Proof Checkpoint Stamp 2 Proof Checkpoint Stamp 1 Proof Checkpoint Stamp 1 Proof Checkpoint Stamp 1 Proof Msg t Msg 2 Msg 1 HMAC HMAC HMAC Nonce Fig 1 High level overview of immutable logging Left depicts an example deployment where an enclaved process generates the logs by capturing message traffi c directly from each source While streaming the log data out to arbitrary storage the data is made immutable by submitting striding checkpoints to the external blockchain comprised of linked integrity proofs that are indexed as checkpoint transactions shown right absolute security of a robotic CPS can not be guaranteed the correctness of such event logs is subsequently tenuous Digital forensic investigations DFIs 7 use digital logs as evidence in post event analysis or in intrusion detection sys tems IDS in electronic devices By continuously broadcasting abridged cryptographic commitments of system state devices are constantly under the sword of Damocles which incentivizes honesty via enforced accountability without relying on specifi c hardware Considering the mass manufacturing restrictions for mobile robots including build of materials serviceability im mense data rates and the utilization of low volume high cost tamper proof storage devices such as Write Once Read Many WORM memory would be fi nancially unprofi table Additionally considering the limitations for mobile robotic platforms such as restricted computational power networking bandwidth on board energy capacity extensive transmission of encrypted logs would be not only technologically impractical but also in violation of international data retention policies As detailed by Veitas et al 8 though shared server centric data retention is favored by OEMs for its straightforward architecture it is also in direct contention with governmental regulatory agencies and privacy advocacy groups Thus our goal is in verifying the integrity authenticity and completeness of robotic event data while under the threat of malicious erroneous insertion omission or replacement To this end we explore the application of a Event Data Recorder EDR based upon cryptographic linked integrity proofs disseminated via distributed ledgers shown in Fig 1 In this work we present Black Block Recorder BBR an IEEE Robotics and Automation Letters RAL paper presented at the 2019 IEEE RSJ International Conference on Intelligent Robots and Systems IROS Macau China November 4 8 2019 Copyright 2019 IEEE approach combining the use of Digital Signature Algorithms DSA keyed hash Message Authentication Codes HMAC and Smart Contract SC via Distributed Ledger Technology DLT to enable tamper evident logging while considering the limited resources available for mobile robotic deployments This is the structure of the paper Section II Related Work discusses the literature on immutable and tamper evident logs distributed ledger technologies and the limits of using existing approaches with robots Section III EDR Roles Requirements and Primitives discusses the details of the properties we want to enforce and what trust settings are addressed with the proposed framework Section IV Approach formulates the integrity proof smart contract and permissioned blockchain architecture as implemented in our framework including design mechanisms and development choices Section V Implemen tation discusses the details of an implementation to evaluate our proposed framework as capable of integrity verifi cation and runtime optimizations for mobile robotic scenarios Finally Section VI Conclusion and Future Work provides a discussion of the work and extensions w r t newer available consensus methods for the practicality and scalability in real world deployments II RELATEDWORK First we give a brief introduction to token based ledgers and their main properties then we discuss in greater detail the concept of distributed ledgers technology DLT immutable logs trusted computing and their relevance to Event Data Recording for autonomous systems A token based blockchain is a peer to peer p2p distributed ledger which derives its security from public key cryptography Each participant in the network has a public address within the Merkle Tree 9 e g derived by the hash of its public key which identifi es the user uniquely among all the other participants Transactions between users are defi ned by pro viding as input the users blockchain addresses the balance transfer and the hashes of the outputs of the last accepted block Candidate transactions are signed and then broadcast in the p2p network and collected by validators that aggregate them in blocks A candidate block is produced when validators mine it by solving the challenge of the consensus algorithm whereupon it will be proposed and added to the chain of previous transaction blocks A proposed fork is only adopted by a validator after it is determined to be the longest chain among the network where all transactions remain valid The security of the approach is assured by the Byzantine Fault Tolerance BFT of the consensus algorithm used and by relying on the diffi culty or inherent cost in subverting the consensus algorithm as a deterrent against malicious actors Readers unfamiliar with this DLT architecture are referred to the seminal work 10 for an approachable introduction A Distributed Ledgers Technology Prior to DLTs horizontally scalable Distributed Databases DDB were commonly used to replicate record state across trusted storage devices However when relying upon CPS infrastructures for data retention auditing the integrity of classical DDB updates in face of transiently available or compromised devices can deteriorate into an under constrained problem Reconstructing postmortem consensus of chronologi cal changes across remaining DDB replications with potentially revoked credentials are classes of issues that can be avoided when disseminating data integrity using DLTs instead As an example Bitcoin 10 provides an alternative to the use of trusted third parties to process and mediate transactions i e the main focal point being the introduction of distributed trust even under mutually distrusting validators The resulting distributed ledger contains an chronological evidentiary trail of consensus that every participant can easily audit As discussed by BitFury and Garzik white papers 11 12 blockchain based ledgers have gained popularity among banks and other fi nancial institutions with the ongoing development of several applications that leverage upon Blockchain s im mutability and consensus to validate transactions However public fi nance blockchains are constrained due to latent limited transaction throughput and scalability due to energy and op portunity costs consumed by traditional Proof of Work PoW 13 consensus To overcome these limitations and enforce enterprise level security mechanisms alternate variants have emerged by defi ning public and private distributed ledgers In public ledgers there are no restrictions on submitting transactions Private ledgers limit those actions to a predefi ned list of entities Ledgers are further classifi ed as permissioned and permissionless In permissioned ledgers vs permission less the identity of peers that act as validators is restricted e g whitelisted public keys Public permissionless ledgers are used for cryptocurrencies like Bitcoin public permissioned ledgers are used to keep control on certifi ed validators private permissioned ledgers work in ways similar to enterprise distributed databases private permissionless ledgers are not possible Even more novel approaches to ledgers have emerged in the Hyperledger Project 14 from Linux Foundation which seeks to improve the performance of the distributed ledgers by creating open source enterprise standard libraries B Immutable Logs Immutable logs require robust tamper proof logging Us ing cryptographic functions we are able to enforce integrity authenticity and non repudiation of the logs entry Several proposals to achieve immutable logs already exist in the liter ature the usual general idea is to use a combination of DSAs and Message Authentication Code MAC to unambiguously validate log entries It is possible to enforce accountability 15 in an heterogeneous distributed environment and reduce the number of trusted devices However the needs of central authorities to store and verify the logs makes it necessary to build an additional chain of trust and deploy a distributed storage system for logs e g distributed databases The use of a distributed versioning implementation such as IPFS 16 can also be a valid option Still the use of Merkle DAG does not incorporate verifi cation mechanisms such as smart contracts which are vital to apply validation logic to the system Following the discussion in II A considering the similarity with Blockchain and its intrinsic security features leveraging on Bitcoin presents an appealing solution 17 Snow et al 18 present how Factom 1 distributes immutable logs on Bitcoin chain using an OP RETURN transaction to store the entry of their client logs Similarly Cucurull et al 19 discuss how at Scytl 2 it was possible to incrementally secure elec tronic voting machine results on Bitcoin blockchain However cryptocurrencies developers regard this as among the more dubious emerging trends in the wild and an abuse of the OP RETURN to piggy back arbitrary data for storage on the Bitcoin Blockchain 20 As discussed by Matzutt et al 21 the impact of this abuse to store non fi nancial content on original cryptocurrency blockchains is unsustainable On the other hand Sutton et al 22 follow the concept of checkpoints presented by Cucurull et al to propose a model using Linked Data to optimize the use of Blockchain by constructing a hashing tree rather than continuously dumping logging hashes into the chain This becomes necessary since the misuse of OP RETURN has several disadvantages either from the protocol point of view discussed above or because of the transaction fees incurred All the transactions that need to be published in cryptocurrency blockchains need to pay a fee that will be burned deducting the limited balance from the account Considering the volatile increase of Bitcoin s exchange rate over the years it s clear that this costly operation is not viable for large scale deployments Another barrier to the use of blockchains for storing im mutable logs is presented by the freshness property of the Blockchain 23 By design Blockchain preserves the order of events i e weak freshness however the accurate time of events i e strong freshness is not guaranteed The work of Szalachowski 24 offers a workaround using a centralized third party however this plays somewhat against our own objectives of distributed trust and scalability Mobile robots may roam autonomously beyond the network range of centralized base stations or any one particular neighbor so any agreed reference to time must arise from a distributed consensus One notable work preceding much of the others thus far using DLT is that of Crosby et al 25 and presents effi cient data structures for tamper evident logging using history trees Although the validation using history trees is effi cient O log2n the runtime time for adding checkpoints is no longer constant O log2n rather than O 1 for hash lists Thus given the lopsided computing resources between robots and off line auditing infrastructure our approach opts for hash lists given the constant overhead in terms of log length while introducing indexing to enable the parallelizion of auditing C Event Data Recorders Event Data Recorders EDR have become prevalent within the automotive industry due in part to regulatory compliance from governmental safety legislation as well as OEM incen tives w r t insurable liability and risk management Reminis cent of Black Box Recorders in aviation EDRs are used to log internal and external vehicular data during deployment such as engine health and status steering and brake operation and 1 2 accident reporting such as obstacle distances or inertial forces from impact Among the list of transportation infrastructure primed to fully incorporate EDR deployments autonomous driving vehicles are perhaps fi rst among them Questions now from both industry and regulatory agencies are being brought forth as per the privacy and security of such EDRs given the pervasive yet critical nature of the data they retain The work by Veitas et al includes a two part series pertaining to these particular issues the fi rst presents Policy Scan 26 a methodology for technology strategy design i e developing concrete actions and products for guiding technology adoption Policy Scan was developed for the purpose of addressing specifi c types of ill defi ned problems in terms of observing analyzing and integrating technology developments with pol icy requirements social governance and societal expectations The second paper 8 applies Policy Scan to the domain of autonomous driving and smart mobility presenting a proposal for making future autonomous vehicles within collaborative intelligent transportation systems C ITS using EDR as more socially acceptable and legally compliant Building upon the above groundwork and also that from Tau rer et al 27 a bio inspired approach to secure data recording for robots we have designed BBR as an EDR implementa tion that conforms to the in vehicle data recording storage and access management requirements as specifi ed while also remaining extendable to general autonomous AI applications using open source robotic middleware and distributed ledger software III EDR ROLES REQUIREMENTS ANDPRIMITIVES Here we formally defi ne EDR systems in terms of the roles requirements and primitives adapted from prior work 8 27 to enumerate our design implementation conformity Boldface terms are later referenced when demonstrating compliance A Obligated Roles and Observing Parties Auditors observing parties called upon to investigate and validate record archives e g Regulatory Agencies or Gov Custodian obligated subject of log content and tasked with log preservation e g Robot or autonomous vehicle OEM Owner mediating party that has a stake in ensuring log integrity authenticity confi dentiality e g End User or Operator Reporter an independent party responsible for faithfully recording events e g Trusted Logger or Recorder Enclave B Recording Storage and Access Requirements R1 Data provision conditions requires consent on behalf of the Owner who transitively controls the log assets tracked R2 Fair and undistorted competition trust should be dis tributed and shared across all validators a k a Custodians R3 Data privacy and data protection the co location of logs external to that of the Custodian mu