// Model Zoo
Pre-trained
Models.
Explore a library of pre-trained models for robotic tasks — optimized for compatibility with RViz, Gazebo, ROS2 simulators, and NVIDIA Isaac Sim.
- Navigation
OL-Nav-Suite v3
Large-scale RGB-D navigation backbone enriched with physics-aware world models and semantic understanding for robust indoor + outdoor navigation.
Params · 1.3BBench · 94.2 SR · HM3DLicense · Commercial - Grasping
OL-Grasp Pro
Comprehensive grasping policy trained on 4.2M demonstrations across 12 embodiments. Includes material properties and failure prediction heads.
Params · 780MBench · 0.91 success · RLBenchLicense · Commercial - Locomotion
OL-Loco Patterns
Multi-terrain locomotion priors with enhanced physics modelling, ground interaction forces, and adaptive gait patterns.
Params · 420MBench · +18% vs SOTALicense · Commercial - Visual Understanding
OL-Vision World
Egocentric vision foundation model trained on 48K hours of synced multimodal capture. Drop-in backbone for VLA and world-model heads.
Params · 2.4BBench · EK-100 SOTALicense · Commercial - Mapping
OL-Map Continuous
Online dense mapping with implicit neural fields, optimized for HMD-class compute and streaming reconstruction.
Params · 340MBench · 2.1cm RMSELicense · Commercial - Motion Planning
OL-Plan Latent
Latent diffusion planner conditioned on hand pose, gaze, and action labels. Ships with RViz / Isaac Sim / Gazebo integrations.
Params · 1.1BBench · 0.88 task SRLicense · Commercial
Need a custom
checkpoint?
Every model in the zoo can be fine-tuned on your embodiment with our custom-collection pipeline. Talk to research.
Pre-trained models for embodied AI and robotics
Model zoo covering navigation, grasping, locomotion, mapping, visual understanding, and motion planning for ROS2 and Isaac Sim.