Nathan Ratliff

Roboticist at NVIDIA. Creator of Riemannian Motion Policies (2018) and Geometric Fabrics (2020-2021) — a framework that derives intelligent robot behavior from the geometry of the space rather than from planning or state machines.

Key Contributions

• Riemannian Motion Policies: Robot behavior defined as second-order dynamical systems coupled with Riemannian metrics. The control law u = M⁻¹·f — metric-weighted acceleration fields — is directly parallel to the ⟨V, G⟩ architecture: M is G, f is ∇V.
• Geometric Fabrics: Strictly generalize classical mechanics to create what Ratliff calls "a new physics of behavior." Obstacles warp the geometry of the space; goals create attractors. Intelligent navigation emerges from the fabric, not from a planner.
• Neural Geometric Fabrics (2023, CoRL): Fabric-structured policies consistently outperform both classical baselines and unstructured neural networks on a 23-DOF dexterous manipulation platform. Structure beats raw computation.

The Key Insight

[REFRAME]

"Motion through a field of obstacles could be just as easy as Cartesian motion as long as we correctly account for how those obstacles warp the geometry of the space." This is the exterior intelligence thesis stated in robotics terms. The difficulty of a task is not intrinsic — it is an artifact of using the wrong geometry. Supply the right metric (G), and the landscape (V) generates behavior that looks intelligent with zero planning.

Geometric Fabrics achieve "intelligent global navigation behaviors expressed entirely as fabrics with zero planning or state machine governance." The robot does not think. It couples to a correctly specified geometric field and behavior emerges.

Role in the Mesocosm

Ratliff provides the engineering proof of exterior-intelligence. His framework was derived independently of Friston's active inference, yet the mathematics is formally identical: both are gradient descent on a Riemannian manifold. Da Costa et al. (2021) showed active inference neuronal dynamics approximate natural gradient descent — the same structure Ratliff builds into robots. Two fields that never cited each other, same control law.

Related

• exterior-intelligence — the unified framework
• karl-friston — independent mathematical convergence
• intelligence-convergence — the argument from convergence