INNOVATION
Innovation in concert
Perception, people, simulation, standards. The deliberate work of shaping the next stage of industrial automation, already running where it matters.
Where the work shows up
Perception — cuVSLAM with NVIDIA
GPU-accelerated visual SLAM that delivers centimeter-accurate localization without fixed infrastructure. Co-authored with NVIDIA, runs on NVIDIA Jetson, and is in production today.
Simulation — Mega Omniverse Blueprint
Full-factory digital twins built into the Idealworks deployment workflow. Customers validate routing, task allocation, and traffic management in simulation before any robot moves on the floor. Faster commissioning, fewer surprises. In production with SICK. Showcased at NVIDIA GTC 2026.
Standards — VDA 5050, MassRobotics, ISO 27001
Idealworks co-led the VDA 5050 3.0.0 release after three years on the core team and 243 pull requests, introducing path sharing and zone-based traffic management. The same posture extends across the standards landscape: contributing to MassRobotics' AMR Interoperability Standard for North American fleet coordination, and to international efforts to align VDA 5050 with broader ISO frameworks. The TUM Cloud Connector, co-developed with TU Munich, keeps the work open and extensible.
Edge compute — Custom IPC with ADLINK
The iw.hub runs on a custom edge AI platform, co-designed with ADLINK and built around NVIDIA Jetson — engineered to the size, weight, and power constraints of a battery-powered AMR running a full shift. SLAM, route planning, depth perception, obstacle detection — all in real time, on the floor, in production. The compute is what makes the perception possible.
Multi-Agent Path Finding — ML-Augmented Fleet Intelligence
Multi-Agent Path Finding (MAPF) has advanced rapidly, but learning-based approaches have remained scattered across the literature and under-explored next to classical search-based methods.
Idealworks' Applied AI Team, in collaboration with the University of Cambridge, co-authored a peer-reviewed open-access survey published in IEEE Access. It's the first to map ML contributions along the full MAPF pipeline — Representation, Planning, and Execution — tracing how learning can augment or replace classical components at each stage rather than grouping work by technique.
The survey consolidates the state of the art and lays out 14 open research questions to guide the next wave of learning-based MAPF.
How we build it
Production first
Every initiative starts from a live production problem. The factory floor sets the brief — the lab answers it.
Open by design
Every robot that speaks the standard language is a robot Idealworks OS can coordinate. The bigger the open ecosystem, the more the orchestration layer can do.
Proven before promoted
What we publish, we ship. Everything on this page exists today, in deployment.
Where innovation leads
Industrial automation has gone through stages. First robots, then fleets, now, orchestration — multiple fleets, multiple vendors, multiple sites, behaving as one system.
Each stage moves repetitive physical work off the floor and pushes skilled work up the chain. Orchestration does the same at a larger scale. Material transport, shuttling, predictable load movement — the work that consumed people for hours every shift — gets handled by the system. The people on the floor run the layer above: orchestrating the operation, troubleshooting, training, making the calls that depend on judgment.
As Physical AI matures and orchestration scales, that shift becomes more achievable — and more important to build deliberately. The companies that get it right won't just have faster operations. They'll have smarter ones across robots, software, and people, in concert.