Industrial AI and Robotics

The Industrial AI and Robotics unit at Obsidian Innovation Institute (Obsidian I²) develops applied methods and deployable prototypes that combine artificial intelligence, human-centred workflows, and robotic/automation integration for industrial environments. The unit prioritises robustness, traceability, and practical validation with industry stakeholders.


Mission

Our mission is to accelerate the adoption of AI-enabled industrial operations by translating research into deployable components and validated workflows. We focus on decision support, human-in-the-loop systems, and integration patterns that can be transferred across industrial sites with minimal overhead.


Core focus areas

We concentrate on the following applied lines of work:

  • Industrial computer vision and perception – inspection and classification pipelines, domain shift robustness, low-/limited-labelled-data strategies, and explainability suitable for operators.
  • Human-in-the-loop decision support – confidence-aware recommendations, escalation logic, operator approval/override flows, and feedback capture for continuous improvement.
  • Evidence capture and traceability – structured recording of inputs, decisions, and outcomes to support auditability, QA sign-off, and reproducible validation.
  • Edge AI and deployment engineering – containerised inference services, performance/latency constraints, and resilient operation under variable conditions.
  • Robotic and automation integration – interface patterns for industrial workcells, safe interaction, and modular integration architectures (including ROS2 where appropriate).
  • Operator experience (UX) for the factory floor – interfaces designed for real operational constraints (time pressure, variability, and cognitive load).

Typical applications

  • visual inspection and defect triage
  • quality decision support and rework guidance
  • adaptive orchestration of inspection-to-action workflows
  • data capture pipelines for continuous process improvement
  • operator-guided robotics and collaborative automation

Methods and capabilities

  • AI/ML prototyping and benchmarking for industrial contexts
  • dataset and taxonomy design (including labelling guidelines and validation protocols)
  • integration of AI services with industrial software stacks and workcell interfaces
  • performance testing and validation under realistic conditions
  • security-by-design and data protection-by-design practices where relevant

Collaboration model

We collaborate with SMEs, integrators, industrial end-users, and research partners through:

  • targeted working groups and technical briefings
  • pilot-oriented co-development of deployable components
  • validation activities focused on measurable operational outcomes
  • dissemination of non-confidential results and good practices

Related units: Artificial Intelligence – Interaction and Multimedia – Cyber Security – High Performance Computing