About

Christopher Aaron O'Hara

AI System Architect, Researcher, and Digital-Twin Engineer

I entered technical work through the places where formal reasoning, physical systems, and human consequences meet: robotics, controls, cyber-physical safety, and resilient infrastructure. My practice focuses on turning advanced research into systems that can be explained, validated, and operated.

Hic Sunt Automata!

Hi, I am Christopher Aaron O’Hara (@ohara124c41), an AI system architect and researcher focused on safety-critical cyber-physical systems, autonomous robotics, and digital-twin platforms. My work spans end-to-end system architecture (requirements to validated deployment), reinforcement learning for decision-making in dynamic environments, and model-based systems engineering for complex, resource-constrained, software-intensive systems. Current interests include human–robot interaction under uncertainty, resilience engineering, certified learning for autonomy, and domain-specific languages for control (IEC 61131-3/PLC) integrated with real-time scheduling and verification.

In recent roles, I have centered my efforts on building the Space Station OS, a modular digital-twin ecosystem (ROS 2 + Isaac Sim + URDF validation with Omnigraph) to prototype GNC/ECLSS/EPS/COMM subsystems and to train/control free-flying robots (e.g., Int-Ball/Astrobee) in microgravity scenarios. This includes interface patterns that transform algorithmic outputs (C++/Python/ROS 2 nodes) into simulator actuation while enforcing physical and operational constraints. Complementary work leads industrial AI initiatives: predictive maintenance for high-throughput bottling (fusion of DPCA with BiLSTM-attention; supervised and semi-supervised RUL/health-status pipelines), anomaly detection with strictly separated holdouts to prevent leakage, and dual-head architectures combining unsupervised deviation scoring with labeled anomaly prediction.

Research contributions include dynamic multi-objective reinforcement learning for hazard-aware navigation (DynaMRPPO) that integrates global–local planning and information-gain objectives; adaptive sensor/filter management with meta-learning and graph attention for energy- and compute-constrained robots; and risk/awareness metrics aligned with human factors for sociotechnical collaboration. Application domains span space robotics (ISS use cases), industrial inspection (e.g., Boston Dynamics Spot risk-aware navigation), and operations optimization in chemical/nuclear-analog environments. Prior collaborative work includes engagements with NASA Ames (airspace separation management), Siemens (COGENT generative/concurrent engineering), ESA, CERN, and Volvo. Most recently, I served as a faculty member/AI researcher at the University of Tokyo (RCAST), working projects at the intersection of digital twins, autonomy, and traditional control systems.

Education comprises a PhD in Aeronautics & Astronautics (University of Tokyo, AI Lab, 2024), an EngD in Software Technology (TU Eindhoven, 2021), MSc degrees in Embedded Systems/Mechatronics/ICT Innovation (TU/e, TU Berlin, NJIT), and an undergraduate background in Philosophy, Science, Technology & Society (Cal Poly Pomona). This interdisciplinary training anchors a practice that unifies formal methods, learning-based control, and human-centered systems engineering.

Selected Projects & Contributions

• Space Station OS (Digital-Twin Platform): Modular ROS 2 package ecosystem with Isaac Sim/URDF validation and subsystem interfaces (GNC/ECLSS/EPS/COMM). Emphasizes extensibility, microservice-style boundaries, and CI-ready design for multi-team collaboration.
• ISS/Int-Ball Microgravity Simulator: Low-fidelity but dynamics-faithful environment for safe navigation, task completion, and certified learning under operational constraints (sensor realism, airflow, microgravity).
• Adaptive Sensor/Filter Management for Space Robots: GAT-based fusion with few-shot/meta-learning to reduce energy/CPU usage while maintaining situational awareness; validated on Astrobee-class platforms.
• Hazard-Aware Navigation via DynaMRPPO: Graph-based, multi-reward RL integrating information gain, terrain handling, and hazard avoidance; improved success rates over standard PPO in complex terrains.
• Industrial Predictive Maintenance: Valve-level DPCA features + BiLSTM-attention fusion for anomaly detection and RUL; rigorous holdout by valve, semi-supervised clustering for unlabeled segments, dual-model decision fusion (regressor + classifier).
• Risk/Resilience Metrics & Human Factors: Definitions and metrics aligning robustness/resilience with operator expectations; sociotechnical framing for human–machine teaming in ISS and process-plant analogs.
• Generative/Concurrent Engineering (COGENT): Methods and tooling for large-scale, multi-disciplinary engineering workflows.

Research & Engineering Interests

Programming language design for controls (DSLs for PLCs, static analysis for timing/safety), RL under constraints, multi-agent coordination, certified learning and gray-box hybrid modeling, system identification for digital twins, and MBSE/SysML-inspired architecture artifacts (IBDs, package/object diagrams) to bridge teams across electrical, mechanical, and software disciplines.

Education

• PhD, Aeronautics & Astronautics (AI Lab) — University of Tokyo, 2024
• EngD, Software Technology — TU Eindhoven, 2021
• MEng, Biomedical Engineering — Colorado State University, 2022
• MSc, Mathematics & Theoretical Computer Science (Embedded Systems) — TU Eindhoven, 2020
• MSc, Electrical Engineering & Computer Science (ICT Innovation) — TU Berlin, 2019
• MSc, Electrical Engineering (Mechatronics) — New Jersey Institute of Technology
• BA, Philosophy, Science, Technology & Society — Cal Poly Pomona, 2015

Programming Languages Spectrum

Preference increases left→right. Depth increases top→bottom. Versions indicate lower bounds.

Notes: ROS 2 is a primary application framework (C++/Python). Familiar with real-time patterns (priority ceiling, rate-monotonic), dataflow models (SDF), and verification-adjacent workflows (tests, property checks) for safety/timing.

Selected Highlights

• Avatarin (space robotics): Adaptive sensor-fusion with GAT + meta-learning achieving double-digit reductions in energy/CPU usage on constrained platforms.
• Tohoku Enterprise (Spot): Graph-based, multi-reward navigation increasing success rate over PPO baselines in cluttered, hazard-rich maps.
• NEC AIOps: LLM wrappers and memory-constrained training pipelines on modest hardware.
• Shibuya Kogyo (RUL/Anomaly): 7-feature DPCA + BiLSTM-attention pipeline; strict holdouts by valve; semi-supervised extensions and dual-head modeling (unsupervised deviation + supervised anomaly).
• NASA/ESA/CERN/Siemens/Volvo: Systems-architecture and AI/controls contributions across safety-critical and industrial contexts.

If collaboration involves digital-twin validation, autonomy under constraints, or PLC/DSL-backed controls with verifiable timing and safety, this is the locus of ongoing work.