Embodied robots are robots that can understand human language instructions and interact in scenarios to complete tasks. They are considered one of the most promising business scenarios in the near future, and many global companies are all competing to announce their progress in this field.
Samsung R&D Institute China – Beijing (SRC-B) has invested years of effort in the field of embodied robots and has continuously achieved top results in international competitions, including human–robot interaction design, embodied navigation, and embodied instruction tracking.
This year, to secure a position in future international competitions, the Intelligent Robotics Committee under the National Computer Commission of China held China’s first embodied intelligent robot navigation and grasping competition. SRC-B was invited to participate in this competition.
The competition required participants to find the target object in a simulated home environment, perform navigation, and grasp operations based on language instructions. After months of fierce competition, SRC-B’s language intelligence team surpassed teams from top companies and universities, and won second place.
Our solution adopts a module-based approach that innovatively incorporates instinctual information during the exploration of unknown environments to facilitate the localization of small objects in complex scenes. In terms of vision, 3D shape completion is utilized to improve the object recognition rate, while data augmentation is applied to target objects to ensure a high recall rate in object recognition. Pretrained grasping technology is employed for the grasping task, ensuring a high success rate in capturing the target object.
This technology is deemed highly suitable for businesses such as nursing homes, kindergartens, office buildings, hospitals, etc., where language comprehension skills are required and simple, interactive, and high-frequency work, such as item handling, inspection, item arrangement, simple device operation, and simple language-defined tasks are required.
Members of the SRC-B Language Intelligence Team