Robota-Embed: Unified VLA Embeddings for Robot In-Context Learning

**Summary:** Neura Robotics announces Neuron AI, a multimodal foundation model designed for Physical AI, integrating 3D vision, touch, audio, robot state, and language with memory to enable action. Based on the PLA (Perception-Language-Action) framework, Neuron AI enhances robotic interaction by incorporating sensory feedback beyond visual input, such as tactile and auditory cues. It is deployed across five robot embodiments and their grippers, with partner customization options. Developed and trained in-house in Metzingen, Germany, Neuron AI is now available via Neuraverse, with access through NEURA Gym.

This paper presents MorphAct, a contextual parameterization framework for robot policies that dynamically adjusts vision-language and action modules based on interaction contexts. The method improves performance across three VLA backbones on LIBERO and Meta-World, achieving gains of up to 16.30%, and shows real-world improvements of up to 14% over a baseline policy.

Robota-Embed is the first vision-language-action (VLA) embedding model that jointly encodes task, execution stage, and future continuation for robots. It achieves 47.33% clip-level and 81.77% episode-level R@1, outperforming action-free Gemini Embedding 2. The model supports training-free and training-based policy conditioning, improving success rates in simulated and real-world tasks, and enables task-local context evolution with reduced policy token usage.