Research and innovation

Norwegian Centre for Embodied AI

Research and innovation

Embodied AI represents a “system-of-sciences” within which the co-design of an artificial body and its intelligence is at the core. It sets the framework for a holistic approach to intelligence and thus calls for a departure from the state-of-the-art in robotics and AI which “exploits” design compartmentalization for development efficiency at the cost of the physical skills and intelligence of robots.

Primary objective

The Norwegian Centre for Embodied AI (NCEI) aims to establish a Universal Science of Embodied AI that enables robots with diverse morphologies across aerial, ground, and aquatic platforms, to achieve nature-like autonomy and performance. The centre will advance fundamental research in key sub domains of Embodied AI while developing a reconfigurable autonomy architecture that integrates perception, learning, control, and computational co design of a robot’s “brain” and “body.” Through this unified and adaptable scientific framework, NCEI seeks to revolutionize how robots are designed and operate in the physical world, positioning Norway as a global leader in physical intelligence and agility research, particularly under the world’s most demanding environmental conditions. The secondary objectives of the project are:

Robot “body-brain computational co-design” – robots should not be ‘merely’ human-engineered but derived by a computational loop driving body evolution as per the fitness of their autonomy policies.
Resilient & Rich Robot Perception – through core breakthroughs in implicit & explicit representations for GNSS-denied navigation and mapping, as well as semantic object-level scene reasoning.
Generalizable Navigation Policies across Embodiments – Investigation of common, re-configurable, architectures for navigation autonomy across flying, ground, surface & underwater robots.
Generalizable Whole-body Manipulation – Ability to conduct dexterous grasping and whole-body manipulation with focus on generalizable yet re-configurable policies for ground/underwater robots.
Common Sense Reasoning and Planning – To solve complex high-level reasoning-to-action plans (e.g., assemble a machine, perform physical repairs) derive a generalist cognition-planning strategy.

WPs and challenges - R&I

Work packages

The research programme of the centre is built around the scientific foundations of Embodied AI. We explore how perception, navigation, manipulation, and reasoning can be coordinated within flexible autonomy architectures that operate safely and robustly across robot types and operational domains.

Impact and methods

Work packages

Our work spans multimodal perception, scene understanding, world models, generalizable policies, whole-body control, and the computational codesign of robot bodies and autonomous capabilities.

By combining simulation-driven evolution, real-world experimentation, and fundamental theory, we aim to define the next generation of algorithms, models, and design principles for physically agile and intelligent autonomous robots.

Our research and innovation activities are structured around a comprehensive programme of Work Packages (WPs) that together advance the scientific foundations, technological capabilities and societal deployment of Embodied Artificial Intelligence.

Simulation as Learning & Testing Basis for Embodied AIs

WP2, led by NTNU: Augmenting NVDIA Isaac Lab & Cosmos for underwater, flying, and walking robot simulation with complex aerodynamic and hydrodynamic effects.

“Body-and-Brain” Computational Co-design

WP3, led by UiO: Evolutionary algorithms are used to computationally evolve robot bodies based on the performance of their autonomy policies.

Perception for Embodied Scene Cognition

WP4, led by NTNU: Novel sensing solutions including Artificial Lateral Lines and nociception, with resilient multi-modal SLAM and semantic scene understanding.

GANPs across Robot Embodiments

WP5, led by FFI, studies Generalizable Autonomous Navigation Policies (GANPs) using World Foundation Models across flying, ground, surface, and underwater robots.

Generalizable Policies for Whole-body Manipulation

WP6 is headed by SINTEF and is dedicated to dexterous whole-body manipulation with generalizable yet re-configurable policies for ground and underwater robots.

Common-sense Reasoning and Planning

WP7 led by SINTEF works on Agentic AI for decomposing complex-high level tasks into actionable robot commands through common-sense reasoning.

Benchmarking Physical Intelligence and Agility

WP8 is headed by UiT and will focus on defining metrics for physical intelligence and agility in open-world navigation and whole-body manipulation tasks.

Field Testing & Hardening of Embodied AI

WP9 is coordinated by OsloMet and will advance real work validation through NCEI challenges and industry-driven field testing campaigns.

Ethics in Embodied Artificial Intelligence

WP10 is led by UiO and will work towards ensuring ethical, safe and responsible development and deployment of Embodied Artificial Intelligence.