MaatInstitute studies how intelligent and autonomous systems can become mathematically verifiable, experimentally reproducible and trustworthy in real-world operation.
Our research asks what a system can prove, who is authorised to act, and what evidence survives contact with the physical world.
How can systems produce machine-checked evidence that their state and computation are correct? Cryptography, zero-knowledge proofs, authenticated data, complexity, theorem proving and hardware acceleration.
How can execution remain trustworthy at the boundary between software, hardware and the physical world? Attestation, trusted execution, runtime enforcement, post-quantum cryptography and safety-critical systems.
How do learned systems generalise from simulation and demonstrations to safe action? Learned control, VLA models, reinforcement learning, manipulation and sim-to-real transfer.
What experimental evidence is sufficient for a trustworthy claim? Closed-loop evaluation, simulation fidelity, provenance, failure analysis and reproducible benchmarking.
We study how cryptographic evidence can run on cyber-physical and autonomous systems across air, ground, aerospace and embodied platforms under strict size, weight, power and cost constraints. The work connects zero-knowledge computation, post-quantum security and hardware acceleration with real-time operation.
Focus areas include MSM and NTT acceleration, FPGA/GPU/ASIC co-design, post-quantum security, authenticated telemetry and verifiable records.
We are interested in how machines sense, move, manage energy and recover from failure in uncertain environments—and how those capabilities can be tested independently in the field.
Focus areas include resilient movement, adaptive bodies, energy-aware control, embodied sensing and reproducible field assessment.
These are application domains, not separate scientific departments. Each one draws on several programmes and gives the research a concrete operational context.
Safety, resilience and predictability of autonomous flight in shared airspace.
Ground machines that drive, deliver and work among people and traffic.
Assurance methods for flight control, avionics and human–machine teaming.
Spacecraft, launch vehicles and high-altitude platforms operating beyond conventional airspace.
Physical machines that perceive, decide and act alongside people.
Independent evaluation of platforms built for contested environments.
Every result states what was observed, inferred and left unresolved, including the validity window and assumptions behind the evidence.
Methods are designed so that a third party can inspect the model, protocol, trace or benchmark rather than relying on institutional trust.
We test stale evidence, runtime substitution, policy bypass, trace loss and disagreement between authenticated sources.
A public record of papers, preprints, posters and research artifacts. Every item uses the same format: status, venue, authorship, a linked title and inspectable source material.
Accepted poster. The OpenReview record is the source of truth for the paper metadata, authorship and publication status.
Additional papers, preprints, datasets, benchmarks and repositories will be added with stable sources, named authors and clear status.
