Physical AI: Neura Robotics and NXAI Combine Research on Robotics

The German robotics manufacturer Neura Robotics and the Austrian company NXAI are working together on memory functions for cognitive robots. As the companies announced on October 20, 2026, they aim to combine Neura Robotics' multimodal foundation model "Neuron AI" with NXAI's recurrent architecture "xLSTM". The goal of the cooperation is to enable robots to handle complex, multi-step tasks while considering previous states and experiences. Additionally, NXAI is expected to provide its own model on the "Neuraverse" platform.

Memory Functions for Multistage Work Processes

The focus of the collaboration is on the advancement of artificial intelligence for physical applications, known as Physical AI. Unlike pure language models, robotic systems must continuously capture, process, and derive actions from information in their environment. They also need to consider which work steps have already been executed and the current state of an ongoing process.

According to the companies, the jointly developed memory functions are intended to cover two areas. The short-term working memory should enable robots to consider current process states and already completed action steps. For a assembly task, a robot could, for instance, recognize whether a component has already been inserted or a specific work step has been performed.

The longer-term episodic memory should also make information from previous workflows available. If a robot repeats a known task, it should be able to draw on stored experiences. The partners aim to create the prerequisites so that robotic systems do not have to process each situation independently of prior workflows.

"xLSTM" complements multimodal foundation model

The technological foundation of the collaboration is the foundation model Neuron AI from Neura Robotics. The model processes different types of information, including three-dimensional image data, tactile sensor data, audio signals, information about the robot's state, and language. These data are linked with a memory function.

NXAI brings the "xLSTM" architecture into the cooperation. This is a recurrent approach to artificial intelligence designed to process information across longer sequences and efficiently update internal states. The architecture is intended to support applications where temporal relationships and previous states are relevant for subsequent decisions.

By combining both technological approaches, the partners aim to link the processing of multimodal sensor data with the long-term storage and use of state information. The joint development is therefore intended to create a technological component for robots that need to coordinate their actions over extended periods.