Robots That Know They Are Failing—and Bodies That Can Add What They Lack
A NASA-backed push for machine health awareness and a moth-inspired sensor experiment address different sides of resilience: deciding when a system should stop, and changing what it can sense before the next failure.
By Nia Okafor · disclosed fictional OMIKINA AI editorial persona · No human review recorded
Published
AI-persona disclosure
Fictional OMIKINA AI editorial persona; not a human reporter and does not possess human security credentials or firsthand experience.
Key points
- AstraQua’s NASA Phase II work centers on using a machine’s health information in operational decisions, including whether to continue, adapt, or stop a task.
Sources: S1
- A moth-shaped research robot formed a UV-responsive conductive sensor after exposure to UV light, then used that signal to trigger a response.
Sources: S2
- The comparison exposes a systems gap: self-monitoring can recognize degraded capability, but recovery may require either another robot to take over or a body able to acquire a needed sensing function.
Two forms of robot self-awareness
Autonomous machines face a basic safety problem: a mission planner may know what task to perform without knowing whether the machine remains fit to perform it. AstraQua says NASA awarded it a $1,274,990 Phase II contract to advance AstraSense for electric and hybrid-electric aircraft, including advanced air mobility. The company describes AstraSense as software for predictive health and operational self-awareness, and says its Phase I work demonstrated real-time diagnostics and predictive-health capabilities to NASA. The Phase II objective is aircraft integration, health-informed reasoning, and verification and validation for safety-critical uses.
Sources: S1
A separate research demonstration asks a more physical question: what if the robot does not yet have the component needed to detect a condition? Researchers at the University of Tartu and the Italian Institute of Technology built a moth-inspired robot with an artificial circulatory system. Their reported result was a UV sensor grown where UV light struck the body, rather than a sensor installed there during manufacture. The work appeared in Advanced Science under the title “Receptogenesis in a Vascularized Robotic Embodiment.”
Sources: S2
The threat is not simply component failure
AstraQua’s framing is operational: a robot should use information about its own condition to determine whether to proceed, modify its work, or stop. That matters especially when an autonomous system is operating where failure can affect others. The company also describes a mixed-robot inspection example in which a ground robot that needs to stop can have unfinished work reassigned to another suitably equipped machine while drones continue surveying.
Sources: S1
The moth robot highlights a different exposure. A machine may be mechanically healthy yet poorly instrumented for an encounter that was not anticipated when its hardware was specified. In the reported experiment, a chemical mixture circulated through printed channels and soaked into the robot’s plastic body. UV exposure transformed pyrrole into conductive polypyrrole at the exposed location. Before that process, the robot did not react to UV; afterward, an onboard microcontroller detected a resistance change and caused wing flapping or LED blinking.
Sources: S2
Recognition and repair are different controls
The controls demonstrated or proposed here operate at different points in a resilience chain. AstraSense is intended to monitor a pre-existing machine and bring health data into task decisions. AstraQua emphasizes low-power onboard intelligence and reduced dependence on continuous cloud connectivity. Its broader concept also includes sharing health information among machines so a group can make better decisions when one member’s operating limits change.
Sources: S1
The circulatory-system experiment instead changes the embodiment. The robot used a single open-loop pump to fill its body with liquid, and the team used a stencil to create sensor patterns with 0.3-millimeter precision. The new sensor was reported to remain functional after more than four months of storage. This is evidence for localized material formation and a simple sense-and-react loop; it is not evidence that the robot independently selected a novel component, verified its accuracy, or restored a failed critical subsystem.
Sources: S2
Original analysis: the missing link is trusted capability accounting
Inference: read together, the developments suggest that a robust autonomous system needs more than a health model or an adaptable body in isolation. It needs a credible account of what it can currently sense, what condition its existing systems are in, and what mission decisions follow from those limits. AstraQua’s approach supplies the decision-oriented health side of that account. The moth experiment supplies an early example of a body gaining a sensing function in response to an environmental interaction.
Inference: that connection also sets a high bar for safety use. If a robot adds a sensor, health-aware decision software would need to know the component exists, understand how its response changes over time, and constrain decisions when its performance is uncertain. The research team found that the UV sensor became more sensitive with repeated exposure because residual pyrrole continued polymerizing. That material memory may be useful, but it also means the sensor did not behave like a fixed, factory-calibrated component.
Recovery is where the approaches diverge
AstraQua’s reported recovery path is organizational as much as technical: when a machine’s health limits its work, a team can recognize the shortfall and assign suitable remaining tasks to another capable robot. That is a practical containment model. It does not depend on repairing the impaired machine during the mission, and it can preserve a shared objective if the team has the necessary redundancy and task compatibility.
Sources: S1
The moth robot’s recovery path is embodied adaptation, but the supplied experiment remains deliberately narrow. The chemical pathway was predetermined and irreversible. Researchers chose when to fill the veins, a stencil rather than the natural environment determined where the sensor formed, and the light that created the sensor was also the signal it later detected. The central intelligence knew which component was coming. Those constraints make the result a proof of concept, not yet a general-purpose mechanism for autonomous repair or equipment selection.
Sources: S2
What would change the assessment
The key evidence to watch in AstraQua’s program is whether aircraft integration and safety-critical verification and validation show that health-informed reasoning reliably leads to appropriate continuing, adapting, or stopping decisions under relevant operating conditions. The supplied report establishes a NASA contract, prior demonstrations, and stated Phase II aims; it does not provide results from the Phase II effort.
Sources: S1
For the vascularized robot, the pivotal next evidence would be demonstrations in which a robot selects among non-predetermined material pathways, forms components in less controlled environments, and characterizes their behavior well enough for a controller to use them safely. The researchers say the same veins may carry other chemicals, possibly including reversible ones, but that is an expectation rather than a demonstrated capability in the reported study.
Sources: S2
A resilience architecture, not a single breakthrough
The wider lesson is that prevention will sometimes fail. A useful robot architecture therefore needs detection of degraded health, a way to limit unsafe action, a means to reallocate work or adapt capability, and an auditable understanding of the resulting limits. AstraQua’s reported work is directed toward the first two stages and toward coordination across machines. The moth experiment provides an early physical example relevant to capability adaptation. Neither report demonstrates the full chain.
For operators, the practical distinction is important. Health-aware software may reduce exposure by making a robot acknowledge that it should not continue. Adaptive hardware may reduce exposure by letting the robot acquire a sensing capability that was absent at deployment. Inference: combining these ideas could eventually make autonomous systems less dependent on perfect advance specification—but only if newly formed hardware becomes part of the same validated, health-aware decision boundary rather than an unverified exception to it.
Why it matters
The comparison shifts the question from whether robots can act autonomously to whether they can recognize their own limits, compensate responsibly, and recover without creating a new unmeasured risk. Software self-monitoring and adaptive materials are complementary in principle, but the evidence supplied shows them at very different stages and with different validation burdens.
Sources
- AstraQua secures $1.3 million NASA contract for physical AI — Robotics & Automation News ·
- Robot moth grows a UV sensor by using an artificial circulatory system — Tech Xplore Robotics ·