Robotics / VLA

NVIDIA's foundation model for humanoid robots. World-model-based VLA enabling whole-body control and human-like manipulation.

Open-source 7B Vision-Language-Action model built on Prismatic VLM and Llama 2. Converts visual observations and language goals into robot actions.

Physical Intelligence's foundation model for robot control. Combines vision-language understanding with dexterous manipulation across diverse tasks.

Google DeepMind's Robotic Transformer 2. Vision-Language-Action model that translates visual observations and language instructions directly into robot actions.

Google DeepMind's Gemini model adapted for robotics. Leverages Gemini's multimodal understanding for zero-shot robot task planning and execution.

Figure AI's VLA model powering their humanoid robots. Combines language understanding with full-body motion planning for household and industrial tasks.

Vision-Language-Action model using LLM backbones for structured robot action prediction. Bridges language models and low-level robot control.

Open-source generalist robot policy from UC Berkeley. Supports multiple robot embodiments and can be fine-tuned for new tasks with minimal data.

Physical Intelligence's latest VLA model with improved generalization. Handles complex multi-step manipulation tasks with fewer demonstrations.

Robotics adaptation of the Flamingo vision-language model. Few-shot learning for robot tasks using language-conditioned visuomotor policies.

Cross-embodiment robotic foundation model from the Open X-Embodiment collaboration. Trained on data from 22 robot types for generalized manipulation.

VLA model with explicit 3D spatial reasoning. Uses depth perception and spatial understanding for more precise robotic manipulation.