Source-linked AI summary
RobResilience: Implementing and Evaluating a Resilience Framework for Cyber-Physical Embodied Systems
Gysella Imrell, Emanuele Miotto, Mahya Mohammadi Kashani, Mauro Conti, Alberto Giaretta
TL;DR
Embodied CPSs need runtime mechanisms that distinguish tolerable disruption and degradation from unsafe states during cyberattacks. RobResilience implements a formal predicate-based framework with IDS-informed compromised-device reasoning and mitigation, and its eight-scenario evaluation confirms consistency with the theoretical definitions.
Problem
Existing embodied-CPS security approaches emphasize detection but lack runtime reasoning about disruption tolerability, degradation bounds, and mitigation during active attacks.
Method
RobResilience implements a formal framework that evaluates δ, γ, and μ over IDS-derived compromised devices in a ROS2/Webots PR2 simulation.
Results
Across eight attack scenarios, the implementation’s runtime behaviour is consistent with the theoretical resilience definitions.
Takeaways & Limitations
The framework preserves the distinction between degraded-but-tolerable operation and non-resilient states while enabling mitigation when feasible.
Takeaways & Limitations
The evaluation bypasses IDS uncertainty, injects symbolic rather than real ROS2 exploits, and models mitigations deterministically.
Abstract
from arXiv · showhide
In embodied cyber-physical systems, active cyberattacks pose an immediate threat not just to data, but to physical integrity and human safety. While existing security approaches excel at detection, they lack the runtime mechanisms to determine whether a disruption is tolerable or if performance degradation remains within safe operational bounds. This gap leaves autonomous systems vulnerable to graceful failure paralysis, where they cannot distinguish between a safe, degraded state and a catastrophic hazard during an ongoing attack. This paper presents RobResilience, an implementation of a formal resilience framework for embodied cyber-physical systems in a Webots simulation environment, using a PR2 robot and ROS2. The framework evaluates three predicates at runtime: tolerable disruption ($δ$), tolerable degradation ($γ$), and mitigation feasibility ($μ$), over a compromised device set derived from IDS confidence scores. When resilience is lost, the framework triggers available mitigation strategies. We evaluate our implementation through eight attack scenarios that systematically cover all possible combinations of the predicate state space, varying attack targets, degradation rates, and mitigation availability. Results confirm that the runtime behaviour of the implementation is consistent with the theoretical definitions.
1 Introduction
RobResilience addresses the under-explored problem of runtime resilience and adaptive mitigation in embodied CPSs during active cyberattacks. It implements and evaluates a formal framework that reasons about tolerability, degradation, mitigation, and operational continuity.
- Motivation: Embodied CPS attacks threaten physical integrity and human safety because these systems combine sensing, decision-making, and physical actuation.Existing work leaves runtime resilience, adaptive mitigation, and graceful degradation comparatively under-explored.
- Contribution: RobResilience implements a formal resilience framework that integrates IDS information, task criticality, embodiment goals, and device dependencies during active attacks.The framework evaluates disruption tolerability, degradation severity, mitigation feasibility, and operational continuity beyond detection alone.
- Implementation: The implementation is deployed as ROS2 software modules in a Webots simulation with a PR2 robot and supports runtime predicate evaluation.The PR2’s mobility and manipulation capabilities enable diverse simulated task scenarios.
- Evaluation: The evaluation covers all realisable combinations of tolerable disruption, tolerable degradation, and mitigation feasibility across critical and non-critical attack targets.The study examines how degradation severity, task-criticality mappings, and adaptive mitigation affect operational continuity.
2 Related Work
Prior CPS security research emphasizes intrusion detection, while related response, FDIR, and formal methods do not provide the same predicate-based runtime resilience mechanism with adaptive mitigation. RobResilience is positioned to fill this gap for embodied CPSs.
- CPS security and intrusion detection: Intrusion detection dominates CPS security research, but detection methods face trade-offs involving known-attack coverage, false positives, and vulnerability to novel tactics.Knowledge-based methods require updated attack dictionaries, whereas behaviour-based methods can generate false positives and depend on training or profiling.
- Runtime response: Existing runtime response work addresses action beyond detection but does not formally evaluate disruption or degradation tolerability over compromised devices.RobResilience evaluates whether compromised systems can continue operating and whether mitigation is feasible.
- FDIR: FDIR assumes known, bounded, and structurally representable physical failures, leaving adversarial intent, task-dependent criticality, and compositional compromise effects unmodelled.These assumptions do not directly transfer from incidental component failure to deliberate cyber compromise.
- Formal and runtime approaches: Formal and robotic security approaches verify behaviour or identify anomalies, but do not evaluate the three resilience predicates with adaptive mitigation over a compromised device set.RobResilience implements the predicate-based framework proposed for embodied CPSs under device-level cyberattacks.
3 Background: The Theoretical Resilience Framework
The theoretical framework models an embodied robot through devices, tasks, goals, criticality mappings, IDS confidence, and runtime predicates. Resilience depends on whether disruption and degradation are tolerable directly or can be restored through feasible mitigation.
- System model: The framework represents a robot as devices supporting tasks and embodiment goals through task- and goal-criticality mappings with levels 0, 1, and 2.Level 0 means none, level 1 important, and level 2 required.
- Compromised devices: IDS confidence scores and device-specific thresholds determine the current compromised device set, with critical devices requiring a lower threshold than base devices.The formal threshold condition is κcrit < κbase.
- Runtime predicates: Tolerable disruption δ is true only when no compromised device is critical for any active task or embodiment-preserving goal.Tolerable degradation γ compares the monotonic performance function ψ(S) against a threshold that is stricter when critical devices are affected.
- Mitigation: Mitigation feasibility μ is true when some mitigable subset of compromised devices can be acted upon so that both δ and γ become true.Available actions include isolation, logic reconfiguration, and reassignment to functionally equivalent non-compromised devices.
- Runtime predicates: Disruption and degradation are independent: non-critical compromises can preserve δ while reducing performance below the degradation threshold and making γ false.Without mitigation, resilience requires both predicates to hold.
- Resilience classification: The resilience posture proceeds from tolerable operation, to recoverability through mitigation, or to safe-state mitigation or halt when neither tolerability nor mitigation is available.The case distinction is specified by Theorem 2.
4 Threat Model
The threat model considers an autonomous embodied CPS attacked through ROS2 message injection against devices supporting navigation, manipulation, and perception. The attacker has partial knowledge, while the IDS and resilience framework are trusted but probabilistic.
- System assumptions: The robot executes tasks autonomously through navigation, manipulation, and perception while continuously evaluating δ, γ, and μ.No human operator is involved in the modeled execution.
- Attacker goals: The attacker seeks to force δ or γ to zero by compromising devices, potentially causing task failure, collisions, or exhaustion of the available time budget.Attack effects may impede navigation, interfere with object acquisition, or corrupt environmental perception.
- Attack model: Attacks are injected at the ROS2 communication layer against one or more devices and remain active until a mitigation subset is deployed.Attack types are selected according to objectives such as tampering with obstacle avoidance.
- Attacker knowledge: The outsider attacker knows available devices and applicable attack types but cannot observe or tamper with criticality mappings, the resilience manager, or IDS outputs.The resilience framework and IDS are assumed trusted and isolated.
- Defenses: The IDS assigns each device a probabilistic compromise-confidence score and detects cyberthreats such as malware signatures, reconnaissance, and denial-of-service flooding.It is not designed to recognize physical wear or non-malicious faults.
5 Proposed Framework Implementation
The implementation translates the formal resilience framework into a continuously running ROS2/Webots system that injects attacks, evaluates resilience predicates, and applies mitigation during PR2 task execution.
- 5 Proposed Framework Implementation: The system implements four modules: simulation environment, coordination controller, attack injection, and resilience evaluation, with runtime data flow illustrated in Figure 2.The controller orchestrates task execution and communication while the resilience module assesses system state and drives mitigation.
- 5 Proposed Framework Implementation: The controller distributes configuration parameters, forwards attack information to both the Attack Executor and IDS, and passes the compromised set to the Resilience Manager for predicate evaluation.The evaluated predicates are tolerable disruption, tolerable degradation, and mitigation feasibility.
- 5 Proposed Framework Implementation: The Task Loop invokes resilience checks during motion primitives and propagates HALTED immediately, while tasks ending in a NOT RESILIENT state are downgraded from DONE to HALTED.This ensures that only genuine recoveries are counted as successful task outcomes.
- 5 Proposed Framework Implementation: The attack injection module publishes symbolic ROS2 messages, parses them into attack records, and applies typed physical effects such as STOP, UNDERSPEED, OVERSPEED, BACKWARD, and GRIP_WEAK.The executor temporarily takes hardware ownership of targeted devices during active attacks.
- 5 Proposed Framework Implementation: The IDS evaluates attacked devices against criticality-dependent thresholds, and the Resilience Manager evaluates δ, γ, and μ at every control step.The IDS maintains the compromised set S, using stricter thresholds for devices critical to the active task or goal.
- 5 Proposed Framework Implementation: The implemented degradation function uses a weighted exponential penalty in which compromised critical and non-critical device counts determine how quickly the system reaches γ(S) = 0.The framework requires monotonic non-increasing degradation, but does not prescribe this specific function; the implementation assigns a steeper early penalty to critical compromises.
6 Evaluation and Results
The evaluation tests eight attack scenarios covering all realisable combinations of tolerable disruption, tolerable degradation, and mitigation feasibility, and compares runtime behaviour with the formal framework. Results show that criticality, degradation severity, mitigation, and IDS-threshold assumptions determine whether tasks remain resilient, recover, or halt.
- Evaluation design: Eight attack scenarios cover all realisable combinations of the disruption, degradation, and mitigation-feasibility predicates under systematically varied attack conditions.Scenarios are grouped by critical-device compromise, degradation tolerability, and mitigation availability.
- Critical-device attacks: A critical-device attack can produce δ=0 while γ=1, showing that disruption and performance degradation are distinct predicates rather than equivalent failure conditions.Scenario 3c sustained ψ≈0.86 above θcrit after the attack, although the resilience requirements still required restoring δ.
- Degradation and failure: High cumulative degradation caused non-resilience even when a later-attacked device was non-critical, because the critical threshold remained active once a critical device was compromised.In the two-wave attack, ψ fell from approximately 0.9 after the critical-device attack to approximately 0.7 after the non-critical attack, below θcrit, and the task halted before baseline.
- Critical-device attacks: Mitigation restored both disruption and degradation tolerability after critical-device attacks, returning the system to a resilient state and enabling task completion.In the reported recovery sequence, the compromised device was removed from S, ψ returned above the critical threshold, and the task finished DONE.
- Limitations: The experiments deliberately set κcrit=κbase=0, guaranteeing that attacked devices entered S but violating the framework’s strict threshold ordering and omitting realistic IDS uncertainty.This isolates predicate and mitigation logic for correctness verification, but assumes effectively perfect detection and weakens evaluation of the criticality-sensitive IDS design.
7 Future Work
Future work should replace simplifying assumptions with realistic threshold learning, probabilistic detection, middleware-level attacks, stochastic mitigation, and physical-robot evaluation.
- 7 Future Work: Realistic threshold values and criticality mappings should replace the currently pre-determined parameters, potentially through machine learning and sim-to-real calibration.The proposed model could use device type, specifications, tasks, and goals to estimate these values.
- 7 Future Work: The evaluation should replace its perfect-detection assumption with a probabilistic IDS and examine how varying κcrit and κbase changes predicate evaluations.The authors hypothesize that lower detection accuracy will make compromised-set composition and resilience classifications less predictable.
- 7 Future Work: Future evaluations should use real middleware-level attacks and realistic mitigation mechanisms instead of symbolic injection and deterministic mitigation.This would enable study of the predicate loop’s stochastic behaviour across repeated runs and robot platforms.
- 7 Future Work: Physical deployment on a real robot is needed to assess timing and noise constraints that the Webots simulation does not capture.Such deployment would test whether the Resilience Manager satisfies real-time requirements in embodied CPSs.
8 Conclusion
Across eight scenarios, the implementation matched the formal framework: criticality governed disruption, mitigation traversed predicate combinations, and disruption and degradation remained distinct. The results also show that cumulative compromise composition can move the system across its resilience boundary, while the study remains simulation-only.
- 8 Conclusion: Across all eight scenarios, the implementation correctly instantiated the theoretical resilience definitions at runtime.The experiments verified the framework’s expected behaviour across the evaluated predicate combinations.
- 8 Conclusion: Attacks on non-critical devices preserved δ=1, whereas every attack on a τ=2 device immediately produced δ=0.Scenarios 1, 2a, and 2b maintained disruption tolerance, while critical-device attacks triggered the disruption path.
- 8 Conclusion: Mitigation restored the affected predicates across evaluated combinations, including simultaneous recovery of δ and γ in Scenario 4b.Scenario 2b recovered γ without changing δ, while Scenario 3b restored δ with γ intact.
- 8 Conclusion: Disruption and degradation were evaluated independently, allowing a system with δ=0 and γ=1 to remain degraded but tolerable before an unrestored critical device caused HALTED.This distinction prevents critical-device loss from being conflated with unacceptable performance, while the formal resilience definition still halts the system to prevent further damage.
- 8 Conclusion: A non-critical device added to an already degraded critical compromise can push ψ below the baseline threshold, making resilience depend on evolving compromised-set composition.The resilience boundary therefore varies with current parameters and the contents of S, rather than remaining a static hardware property.
- 8 Conclusion: The evaluation used only synthetic Webots experiments with a virtual PR2 and no human subjects or real systems.The study’s data and configurations were openly released.