PaperlayerView exact revision on arXiv

Source-linked AI summary

Mission-Aware Attestation Envelopes for Time-Critical Autonomous Action: A Hardware-in-the-Loop V2I Study

Dimitrios Nikou, Nikolaos Kekatos, Sophia Petridou, Stylianos Basagiannis

arXiv:2610.08771v1cs.CRcs.ROcs.SYeess.SY

TL;DR

Privileged V2I actions depend on integrity evidence, but the evidence ages while the decision takes time and the physical system imposes a deadline. The paper models these constraints together and evaluates their interaction on a hardware-in-the-loop platform, finding that freshness and attestation interval materially shape where decisions remain actionable.

  • Problem

    Integrity verdicts are commonly treated as predicates, leaving uncoupled whether evidence remains fresh and a decision arrives before a physical action deadline.

    Source: §1 · p. 2

  • Method

    The paper formulates mission-aware attestation as a four-outcome runtime assurance contract and combines simulator-derived mission state with measured TPM, IMA and verifier behaviour in a hardware-in-the-loop evaluation.

    Source: §1 · p. 3

  • Results

    A fivefold latency scaling removes nineteen feasible operating points, versus twenty-two when the attestation interval changes from two to ten seconds; the security-blind envelope is larger, with the difference concentrated in evidence age.

    Source: §6 · p. 20, §8 · p. 23

  • Takeaways & Limitations

    Attestation interval is a directly configurable deployment parameter, and freshness bounds cannot be chosen from security requirements alone when they make every late decision also stale.

    Source: §1 · p. 3, §8 · p. 23

    The tamper experiment establishes detection only for modifications to files covered by the measurement policy, with a decision process that is not itself adversarial.

    Source: §4.1 · p. 11

Abstract

from arXiv · show

An autonomous system that asks for a privileged physical action is usually gated on integrity evidence: a platform proves what it is running, and the request is granted or refused on that basis. Such a gate is normally treated as a predicate, yet the evidence behind it has an age, the decision that consumes it has a latency, and the physical system that waits for it has a deadline. We formulate mission-aware attestation as a runtime assurance contract that holds only when integrity is valid, the evidence is fresh enough, and the decision completes inside a budget derived from the current physical state. The contract yields four operational outcomes where a binary gate yields two, separating a refusal caused by tampering from one caused by stale evidence and from one caused by a late decision. We evaluate it on a hardware-in-the-loop vehicle-to-infrastructure platform: a driving simulator supplies the physical state and the authorisation deadline, while a microcontroller on-board unit and a TPM-backed roadside unit running Linux integrity measurement supply the assurance evidence. A security-blind model admits the whole operating space and a hardware-informed one three quarters of it, and every point it refuses fails the freshness margin rather than the response margin. Moving the attestation interval across the range the verifier permits costs about as much as a fivefold scaling of the latency distribution, and the interval is directly configurable, which makes it the immediately actionable deployment parameter. If the freshness bound does not exceed the authorisation budget, every late decision is also stale and lateness becomes unobservable, so the attestation interval and the freshness bound cannot be chosen from security requirements alone.

1 Introduction

The paper argues that integrity evidence must be judged by its freshness and by whether the decision arrives before a physical deadline, not merely by whether the integrity verdict passes. It proposes a runtime assurance contract and a hardware-in-the-loop method for evaluating where that contract is actionable.

1 Introduction

  • V2I lets vehicles request privileged changes to infrastructure over a shared channel, making remote authorisation a security and timing problem.Communication delay is already an established input to signal-preemption performance.Source: §1 · p. 2
  • Prior work recognizes freshness and timeliness separately, but does not couple the verdict to a physical action deadline set at runtime by requester position and speed.The paper frames this missing coupling as whether the verdict remains decision-valid for the action.Source: §1 · p. 2
  • Four contract outcomes distinguish integrity violations, stale evidence and late decisions, requiring integrity validity, freshness and timely completion together.The response and freshness margins are checked alongside integrity; only their conjunction permits the action.Source: §1 · pp. 2–3
  • A hardware-in-the-loop method combines simulator-derived physical state with real TPM, integrity-measurement and network behaviour to map an actionable operating region.The simulator is not synchronised with or calibrated against a physical intersection, so the authors do not describe it as a digital twin.Source: §1 · p. 3
  • Attestation interval and decision latency move the feasible envelope comparably, and the interval can be changed without hardware changes.The evaluation compares a security-blind envelope with one informed by TPM, IMA and verifier behaviour.Source: §1 · p. 3

2 System Model and Assurance Problem

The system model describes a time-critical V2I authorisation loop and the trust, threat and timing assumptions around it. It motivates a richer assurance contract because a binary gate cannot distinguish several materially different refusal cases.

2.1 Operational V2I Scenario

  • A vehicle requests signal preemption, and the infrastructure must decide before arrival; trigger distance and speed determine the authorisation budget.The model generalises to privileged-access requests where a physical process sets how long the decision may take.Source: §2.1 · p. 4

2.2 System Model

  • The model separates requester identity checks from periodic platform-integrity checks, whose evidence age grows between attestations and during decision latency.It represents the system as requester, decision point, evidence source and physical plant, with request and decision events carrying state, time and outcome.Source: §2.2 · p. 5

2.3 Threat and Trust Model

  • The evidence source and physical-state reports are trusted, while the decision point is not; an adversary may modify its files or generate arbitrary channel traffic.The model excludes forging integrity evidence, compromising the evidence source, and several physical and supply-chain attacks.Source: §2.3 · p. 6
  • Channel traffic can exhaust the response margin and deny service without defeating identity or platform-integrity checks; the measured study does not exercise this capability.The physical-state-derived budget is fixed and not itself attackable.Source: §2.3 · p. 6

2.4 Limitations of a Binary Gate

  • A binary gate reports identity and platform-integrity failures, but cannot distinguish a correct decision arriving too late from evidence that is too old to trust.Under a fail-secure policy, a late decision preserves infrastructure safety but the requester loses the service; periodic evidence may be up to T old, or Tmax within the verifier's permitted window.Source: §2.4 · pp. 6–7

3 Mission-Aware Attestation Contract

The section defines mission-aware attestation as a contract for deciding whether integrity evidence is actionable within a physical deadline. It distinguishes freshness from response timeliness and specifies how their interaction affects refusal outcomes and runtime monitoring.

3.1 Authorisation Budget, Evidence Age and Margins

  • The request records time and physical state, which sets D(dr, vr) = dr/vr as an upper bound on the authorisation budget; evidence age and decision latency use different origins.The freshness bound Fmax is distinct from the verifier liveness threshold Tmax; a negative response margin means a correct verdict arrived too late to use.Source: §3 · p. 7, §3.1 · p. 7, Figure 2 · p. 8

3.2 Contract and Outcome Semantics

  • The contract permits an action only when integrity passes and both freshness and response margins are nonnegative, assigning distinct outcomes by integrity, freshness, then timeliness.When both margins are negative, the precedence rule reports DENY-STALE rather than DENY-LATE.Source: §3.2 · pp. 7–9

3 Mission-Aware Attestation Contract

  • The attestation envelope is the operating-space subset where evidence satisfies the contract and is actionable.The section frames the contract as separating three requirements that a binary gate conflates.Source: §3 · p. 7, §3.2 · p. 8

3.3 Interaction Between Freshness and Timeliness

  • With periodic held evidence, Fmax ≤ D(xr) makes every late decision stale, so DENY-LATE is masked; it becomes reachable only when Fmax > D(xr).The evaluated policy sets Fmax = D(xr); on-demand refresh breaks the held-evidence premise and makes DENY-LATE observable at any Fmax.Source: §3.2 · p. 8, §3.3 · p. 9

3.4 Runtime Monitor Realisation

  • The monitor carries an absolute deadline as data and checks recorded terms with arithmetic and event ordering rather than a variable metric interval.No engine cross-check is reported; discharging the obligation on a temporal engine would validate the encoding, not the contract.Source: §3.4 · p. 10

4 HIL Platform and Experimental Method

This section describes a hardware-in-the-loop method that combines simulator-derived mission state with measured attestation behavior, then specifies how timing, evidence age, contention, and operating factors feed the envelope inputs.

4.1 Hardware-in-the-Loop Platform

  • CARLA supplies simulated vehicle state while a physical ESP32 sends admission credentials to a Fedora RSU running TPM 2.0, IMA, and Keylime.Edge checks establish admission rather than cryptographic identity; infrastructure verification checks measured files against policy.Source: §4.1 · pp. 10–12
  • The measured agent, verifier, and decision logic share one host, so the platform does not establish protection against a compromised decision consumer.A separate verifier host or independently gated signal actuation would place the verdict consumer outside the attested boundary.Source: §4.1 · pp. 11–12

4.2 Timing and Evidence-Age Instrumentation

  • Durations use same-clock readings at each observation point; end-to-end latency is conservative because it starts before request time, while evidence age comes from verifier records.Only end-to-end and status-query durations are separated; REST replaces per-request CLI subprocess overhead.Source: §4.2 · pp. 12–13

4.3 Experimental Factors and Configurations

  • Experiments sweep trigger distance, speed, evidence-age phase, and verifier-permitted attestation intervals; phase is normalized to interval length.Contention is reported as measured throughput; the saturating condition sustained approximately 17 thousand messages per second.Source: §4.3 · p. 13, §4.2 · p. 13

4.4 Metrics and Statistical Analysis

  • Latency is summarized by mean, median, p95, and maximum observed, while contract-satisfaction fractions are point estimates rather than certified bounds.Wilson score intervals are retained but not reported; no execution-time or network bound is established.Source: §4.4 · p. 14

5 Results

The results show that incorporating measured assurance latency and evidence age narrows the feasible operating envelope, with freshness accounting for the observed infeasible points. Sensitivity and failure-mode analyses show how interval choice and the contract’s outcome categories shape what the evaluation can distinguish.

5.1 Assurance Latency and the Effect of Contention

  • Under contention, the mean rises to 538.7 ms, a 27.4% increase; rank tests support an end-to-end shift (p = 2.5 × 10^-3), but not an attestation-component shift (p = 0.88).The loaded arm had seven trials; its mean is only a fragile point estimate, and the retained statistics cannot establish the shift’s magnitude.Source: §5.1 · pp. 14–15, Table 4 · p. 15

5.2 Constructing the Attestation Envelope

  • 76.7% of operating points remain feasible under AHI, while Aideal admits all by construction; losses cluster where trigger geometry is tight.The 95% satisfaction criterion is empirical, not a confidence statement about the underlying success probability.Source: §5.2 · p. 16, Table 6 · p. 17
  • The envelope boundary is D − ϕT ≥ Q(q), so feasibility depends on the measured latency quantile together with evidence age.The closed-form boundary reproduces the sixty-point grid classification.Source: §5.2 · p. 16
  • At 110 km/h, the worst-phase trigger requirement is 78 m at the configured period, versus 17 m with always-fresh evidence and 323 m at the tolerated maximum.The required geometry is derived from the attestation interval and decision cost.Source: Table 5 · p. 16, §5.2 · p. 17

5.3 Sensitivity to Attestation Interval and Decision Latency

  • Nine points fail from evidence age alone at every latency scaling, while combined age and latency failures increase from five points to 24 under fivefold slowdown.No sampled decision is late on the evaluated grid; extending it below 20 m would exercise the response margin.Source: §5.3 · p. 18
  • Across the configured 2-to-10-second range, the interval costs 22 feasible points, comparable to 19 lost under fivefold latency scaling (ratio 1.2).At the tolerated maximum, every point lost beyond the nominal period fails on freshness, with none lost on latency.Source: §5.3 · pp. 17–18, Table 6 · p. 17, Table 7 · p. 18

5.4 Failure-Mode Discrimination

  • The contract attributes hardware-executed identity and policy-file integrity refusals separately from a derived stale-evidence refusal; DENY-LATE is absent under the evaluated freshness policy.Two conditions are exercised on hardware, one is derived, and one is unreachable at these geometries.Source: §5.4 · pp. 19–20, Table 8 · p. 19

6 Discussion and Threats to Validity

The discussion limits the claims to a single platform instantiation and identifies constraints in the simulator, transport, measurements, and untested scenarios. It also compares the envelope’s sensitivity to latency scaling and attestation interval changes.

6 Discussion and Threats to Validity

  • A fivefold latency scaling removes nineteen feasible points, while moving the attestation interval from two to ten seconds removes twenty-two.This compares sensitivity only; the study does not attempt an attack or claim attacker effort or realisability.Source: §6 · p. 20

6.1 Threats to Validity

  • The numerical boundaries apply to one intersection, one on-board unit, one verifier, and one vehicle at a time.The formulation and joint evaluation of evidence age and response latency against a mission-state-derived deadline are presented as generalisable.Source: §6.1 · p. 20
  • The simulator is not calibrated against a physical intersection, and vehicle dynamics use a constant-speed approach.CARLA is a simulator, not a digital twin.Source: §6.1 · p. 20
  • MQTT over IP measures shared-segment contention, not V2X radio contention, so its magnitude should not be transferred to a radio setting.The on-board unit’s uplink acknowledgement round trip has no counterpart in a broadcast radio layer.Source: §6.1 · p. 20
  • Fifty unloaded trials retain samples, but seven loaded trials retain only summary statistics, supporting a contention effect rather than its magnitude.Larger, equal-scale arms would be needed to estimate that magnitude.Source: §6.1 · p. 21
  • The unloaded summary line conflicts with recomputation from the fifty rows, so the paper reports recomputed means and treats the samples as primary.The reported summary line gives 421.77 ms end-to-end and 29.17 ms attestation means; recomputation gives 422.83 and 33.30, while minima and maxima agree.Source: §6.1 · p. 21
  • Evidence age is resolved only to one-second granularity from the verifier timestamp, so the envelope’s phase axis is derived from the configured interval rather than measured per request.Direct per-request resolution would require an agent-side timestamp.Source: §6.1 · p. 21
  • The study does not demonstrate a denial attack, evaluate adaptive refresh, decompose latency by term, or instantiate concurrent requesters.Contention detectably slows the loop, but the measurements do not establish by how much; the adversary considered cannot forge identity or platform integrity.Source: §6 · p. 20, §6.1 · p. 21

7 Related Work

Related work treats attestation freshness, constrained-device evidence, timeliness, vehicular communication, and runtime verification as distinct strands. This work connects those concerns by evaluating an attestation verdict against a physical-state-derived authorisation deadline.

Attestation architectures and evidence freshness

  • Prior attestation work treats freshness as a configured policy quantity, while this work derives recency requirements from the process controlled by the verdict.Freshness and expiry are established concepts in remote attestation and the RATS architecture.Source: §7 · pp. 21–22

Attestation of constrained devices

  • Constrained-device attestation studies what evidence devices can produce and at what cost; this work asks how to evaluate the verdict after evidence exists.Examples include collective attestation and control-flow attestation.Source: §7 · p. 22

Timeliness as a security requirement

  • ARI treats the execution as the attested object, whereas this work treats the platform as attested and times the decision consuming its verdict.Unlike schedulability approaches that assume execution-time bounds, this study reports measured distributions rather than certified bounds.Source: §7 · p. 22

Vehicular communication, availability and preemption

  • Vehicular communication research models delay and preemption constraints; the paper uses flooding as a stressor and derives its deadline from the authorisation decision.The deadline is not a preemption deadline.Source: §7 · p. 22

Runtime verification for cyber-physical and edge systems

  • Runtime-verification work studies monitoring specifications and monitor placement, while this paper’s contribution is the envelope induced when the deadline varies with physical state.The related work spans temporal monitoring, edge topologies, co-simulation, and digital twins.Source: §7 · pp. 22–23

8 Conclusion

The paper frames attestation as a contract over evidence age, decision latency, and a physical deadline, then evaluates the resulting envelope on a hardware-in-the-loop platform. Its findings show that freshness and response margins interact and that neither the attestation interval nor freshness bound can be selected from security requirements alone.

8 Conclusion

  • Below a threshold on the freshness bound, every late decision is also stale, making lateness unobservable.The four outcomes distinguish failures that a binary gate reports identically or not at all.Source: §8 · p. 23
  • A security-blind model admits a larger operating envelope than a hardware-informed model, with the difference concentrated in evidence age rather than decision latency.A fivefold latency scaling removes nineteen feasible operating points, compared with twenty-two when the interval moves from two to ten seconds.Source: §8 · p. 23
  • Neither the attestation interval nor the freshness bound can be chosen from security requirements alone; both must be assessed against a mission-dependent envelope.The envelope concerns when evidence age and the time to consume it remain within the mission-dependent decision horizon.Source: §8 · p. 23
Loading 2610.08771v1…