SerpApi.Org
Chinese robotics company Unitree has unveiled UnifoLM-OminiA-0.3, a unified AI model for humanoid robots.
The model enables real-time omni-modal interaction, reasoning, dialogue, and whole-body mobile manipulation in a single system.
The company says the model is aimed at home-care and wellness applications, allowing robots to autonomously carry out tasks such as tidying rooms, opening drawers, and assisting people around hospital beds while responding to voice, visual, and environmental inputs.
Recently, Unitree unveiled a low-cost upper-body humanoid robot, priced from 26,900 yuan ($4,290), aiming to make humanoid robotics more affordable and accessible.
Smarter humanoid control
UnifoLM-OminiA-0.3 is a new embodied AI model designed to enable humanoid robots to perform a wide range of household, wellness, and care-related tasks using a single control system.
The Chinese robotics company says the model combines real-time reasoning, dialogue, omni-modal perception, and whole-body mobile manipulation, allowing robots to understand spoken instructions, interpret visual information, navigate their surroundings, and interact with objects without relying on separate task-specific models.
According to Unitree, the unified architecture allows a robot to process multiple sources of information simultaneously. Instead of switching between different AI systems for speech recognition, object detection, navigation, and manipulation, UnifoLM-OminiA-0.3 integrates these capabilities into one model. This enables the robot to receive a verbal request, understand its environment, plan an action, move to the required location, and complete the task autonomously, reports Rocking Robots.
The company demonstrated the model through a series of household and care-related activities. In the video, a humanoid robot picks up a cushion and places it on a sofa, identifies colours, counts medicine boxes, retrieves a requested medicine package from a shelf, and sorts clothes into a laundry basket. The robot also loads a plate into a dishwasher and adjusts a hospital-style bed.
One demonstration highlights the model’s ability to respond during task execution rather than only before an action begins. While the robot is adjusting the bed, the user instructs it to stop midway through the task. The robot immediately responds to the interruption and halts the operation, illustrating Unitree’s focus on continuous human-robot interaction and dynamic task adaptation.
Unified manipulation model
Although many of the individual tasks shown in the demonstration have previously been performed by humanoid robots from other companies, Unitree says the significance of UnifoLM-OminiA-0.3 lies in its ability to coordinate the entire workflow through a single AI model.
The system combines language understanding, visual perception, object recognition, decision-making, navigation, and full-body motion control within one framework. The company says the model is designed to execute tasks autonomously while maintaining stable performance even when interrupted or faced with unexpected changes, reports Techiexpert.
The approach reflects a broader trend in embodied AI, where robotics developers are increasingly integrating vision-language models directly with robot control. Rather than programming separate routines for every object or activity, these unified systems aim to generalize across different environments and tasks. Such flexibility is considered particularly important for homes, hospitals, and care facilities, where objects are frequently moved, instructions can change, and people may interact with robots while they are working.
UnifoLM-OminiA-0.3 is the latest addition to Unitree’s broader UnifoLM embodied intelligence programme. In 2025, the company introduced UnifoLM-WMA-0, an open-source world-model-action framework designed to help robots understand physical interactions and predict environmental changes. This was followed by UnifoLM-VLA-0 in early 2026, a vision-language-action model for general-purpose humanoid manipulation released with open-source training code and model weights.
Unitree has also published an expanding collection of whole-body teleoperation datasets captured from humanoid robots performing everyday activities. Many of the household tasks featured in these datasets, including cleaning and object handling, also appear in the UnifoLM-OminiA-0.3 demonstration, highlighting the company’s ongoing effort to develop increasingly capable general-purpose humanoid robots for real-world environments.
