AgiBot releases humanoid manipulation dataset to enable large-scale learning

Robot training, especially for emerging humanoids for diverse environments or applications, requires a lot of high-quality data. AgiBot today announced AgiBot World, a large-scale robotic learning dataset. The developer of humanoid robot technology said it designed the dataset to advance multipurpose robotic policies.

AgiBot World is an ecosystem that includes foundational models, standardized benchmarks, and a collaborative framework aimed at democratizing access to high-quality robotic data. AgiBot also said its dataset provides an opportunity for academia and industry to collaborate, which could result in a leap forward toward universal, adaptable robotic intelligence.

Many existing robot learning benchmarks face significant limitations when addressing real-world challenges, the Shanghai, China-based company noted. These issues primarily stem from low-quality data and restricted sensing capabilities, AgiBot said. This results in benchmarks that are often constrained to short-horizon tasks within controlled environments.

Such limitations hinder progress toward generalizable and robust robotic systems capable of operating effectively in unstructured, dynamic real-world settings, according to the company.

AgiBot World aims to help embodied AI collaboration

With more than 1 million trajectories from 100 robots, AgiBot said its model offers a high level of diversity and complexity. The dataset spans more than 100 real-world scenarios across five target domains, enabling it to tackle fine-grained manipulation, tool usage, and multi-robot collaboration.

The company said it designed these scenarios to reflect the nuanced demands of real-world robotic applications.

AgiBot World offers multimodal hardware including array-based visual tactile sensors, durable hands with six degrees of freedom (DoF), and mobile dual-arm robots with whole-body control. The company claimed that these features will open new frontiers for research in areas such as multimodal imitation learning, multi-agent collaboration, adaptive manipulation, and beyond.

AgiBot said it aspires to transform large-scale robot learning and advance scalable robotic systems for production. The open-source platform invited researchers and practitioners “to collaboratively shape the future of embodied AI.”

Register today to save 40% on conference passes!

AgiBot also offers humanoid platform

Established in February 2023, AgiBot said it is dedicated to “in-depth” artificial intelligence and robot fusion. The company has raised funding in an angel round and A1 to A4 rounds.

In addition to AgiBot World, the company offers AgiBot A2, a second-generation, general-purpose humanoid prototype. AgiBot said this robot is suitable for marketing and customer service, exhibition hall presentations, supermarket guidance, front desk reception, and business inquiries. 

The robot stands at 169 cm (66 in.) tall and weighs 69 kg (152 lb.). It has a runtime of two hours and has more than 40 DoF, explained AgiBot. The company said it supports customized exterior options for the robot to better meet diverse user scenario requirements.

China hopes to gain traction in the humanoid market

The humanoid robotics space has been heating up in recent years, and China is determined to keep up with the rapidly developing market. In November 2023, Beijing’s Ministry of Industry and Information Technology (MIIT) published national goals for the development of humanoid robots. The MIIT predicted that humanoids would become a disruptive technology in the coming years and said that China should prepare to mass-produce such robots by 2025.

In October 2024, the country launched a data-sharing initiative for its domestic humanoid companies to accelerate innovation. The project is being spearheaded by the National Local Joint Humanoid Robot Innovation Center and various Chinese humanoid companies.

The National Local Joint Humanoid Robot Innovation Center (NLJIC), China’s first public platform dedicated to this type of robot, is also building a training ground for manufacturers to collect high-quality data and promote industry-wide standards.

Despite these efforts, humanoids created by Chinese companies haven’t found the same traction as their international counterparts. Last week, Figure AI announced that it has delivered its Figure 02 systems to a paying customer. Engine AI recently introduced its own model.

Other companies that have conducted commercial trials of humanoid robots include Agility Robotics, Apptronik, Boston Dynamics, and Tesla. Their systems are being tested in manufacturing and warehouse settings by Amazon, GXO Logistics, Mercedes-Benz, and Schaeffler.