01
Cryptographic Machine & Agent Identity
Device-bound keys, hardware roots of trust (TPMs, Secure Elements), certificate lifecycle management, and identity federation across edge, cloud, and on-chain systems.
A growing class of autonomous systems - vehicles, sensors, drones, robots, and AI agents - increasingly provides services, consumes services, and settles transactions with one another. These Machine Services introduce new requirements for security, identity, and data architectures: Who is the acting entity? Can the transmitted data be trusted? Is one machine authorized to act on behalf of another?
FUTURLAB studies the cryptographic, protocol, and systems foundations required to make identity, provenance, and integrity verifiable across machine-driven value chains - at a scale and speed where human verification is no longer feasible.
Device-bound keys, hardware roots of trust (TPMs, Secure Elements), certificate lifecycle management, and identity federation across edge, cloud, and on-chain systems.
Signed and chained data streams, tamper-resistant telemetry, and audit trails for AI, compliance, and automated settlement.
End-to-end encrypted peer-to-peer protocols, mutual authentication, and scalable networking for large machine fleets.
Contract, settlement, and incentive mechanisms for Machine Services at the intersection of identity, decentralized settlement, and real-world service delivery.
Hardening LLM- and MCP-based architectures against prompt injection, tool poisoning, and attacks in multi-agent environments.
Privacy-by-design, data residency, regulatory auditability, and decentralized architectures for critical infrastructure and industrial IoT.
RoboPay is payment infrastructure for AI agents in the physical world - a trustless payment layer for autonomous machines (robots, drones, delivery vehicles) that confirms real-world task completion (e.g. a physical delivery) and releases payment from a Solana escrow smart contract, with no human in the loop.
Industrial robotics is evolving from centrally controlled single machines into semi-autonomous fleets that accept tasks, consume resources, and deliver services. Economically they remain dependent: every settlement still needs a human to confirm it. The missing piece is not payment technology but evidence - machine-provable proof that a specific device actually performed a specific service, on untampered measurements. And as AI inference moves onto the device, the raw data never leaves it - so after-the-fact verification becomes structurally impossible. The trust anchor must move onto the machine itself.
A small delivery robot (PiCar-X) uses an AI model to visually confirm a physical delivery (camera image + distance sensor) before autonomously releasing payment from a Solana escrow smart contract - proven with a working prototype and live Devnet transactions.
The main reasoning loop - multi-step tool use across the robot's sensors: look at the camera, check distance, decide, confirm the delivery.
An independent “second-opinion” service demonstrating the x402 machine-payment protocol combined with a separate AI check before release.
A core research question: what should count as proof that a machine really completed its task? RoboPay benchmarks three trigger mechanisms for machine-to-machine settlement:
Location-based proof - the machine reached the delivery coordinates.
Scan-based proof - a code confirmed at the point of handover.
Vision-based proof - an agent visually verifies the delivery actually happened.
How can an edge-AI model's output be attested so that model version, executing device, and result can later be conclusively linked?
How can proof and settlement continue locally during network loss and reconcile consistently on reconnect?
Which governance and authorization models let a machine trigger transactions under delegated budgets - including revocation and liability?
What compute, energy, and latency does continuous attestation impose on constrained hardware - and where is the limit of practicality?
Why it's new: the contribution is not a single cryptographic primitive but the closed chain from sensor to settlement - tested under real industrial field conditions rather than in the lab. Unlike overlay networks (WireGuard, Tailscale, Nebula) that secure the transport, RoboPay secures the origin and integrity of the claim itself.
Research paper: robopay-group.gitlab.io ↗ · Follow: @alexandrastaex · @Staex_io
Attestation architecture, security analysis, and evaluation methodology. htw-berlin.de ↗
Protocol stack, edge software, and field infrastructure for machine networks. staex.io ↗
Real robotic fleets for field deployment and industrial evaluation.