Increasingly, companies with workplaces where large numbers of people are employed in manual labour, such as factories and warehouses, face a future full of questions. There’s the issue of recruiting enough workers; the ever spiralling costs of employing those workers; and the need, in an era of growing industrial automation, to constantly improve efficiency and safety.
Most current robots, however, are hampered by common shortcomings: they can be cumbersome in their movement, for example, or incompetent at understanding or navigating complex environments, or limited in their ability to execute tasks accurately.
Riding to the rescue is a new AI robotics platform developed by CUHK’s InnoHK Hong Kong Centre for Logistics Robotics (HKCLR). The first such system to be developed in Hong Kong, it consists of two components: the M1 quadruped robot and the dual-arm embodied AI system powered by a Vision Language Model (VLM).
The team started by considering the issues facing the industrial and logistics sectors, says Professor Liu Yunhui, Choh-Ming Li Professor of Mechanical and Automation Engineering in CUHK’s Faculty of Engineering and Director of HKCLR. “We began exploring innovative solutions, including robotics and embodied AI. In this context, quadruped robots emerged as promising tools, potentially addressing these issues by improving last-mile delivery, warehouse automation and inspection, while also helping to reduce costs and enhance workplace safety. While several quadruped robots have come into the market, challenges remain in payload, motor overheating, stable locomotion and autonomous navigation in natural environments, to which we are aiming to develop new solutions.”
M1: Wheeled quadruped robot
M1 stands out as the world’s first wheeled quadruped robot with parallel joints. This innovative design enhances its flexibility and compactness, allowing it to adapt seamlessly to challenging outdoor terrains while increasing its payload capacity.
M1, meanwhile, is exceptionally talented at perceiving and understanding its environment, making appropriate decisions based on that information. It’s also extremely skilled at navigating tough terrain, whether that be stairs, slopes, grass, ice or other uneven surfaces. One of its standout features is the Butterfly Slope Assault, which enables the robot to navigate through narrow spaces with ease. It can even scale obstacles up to 100 cm in height and constantly improve the way it moves through reinforcement learning.
M1, which is 950 mm long, 450 mm wide 570 mm high, is primarily made from aluminium alloys, materials that offer a winning combination of lightness and durability. The innovative differential drive system that powers it helps increase its payload capacity to between 15 kg and 40 kg, depending on the model. It can transport square objects on its back, while a track and basket attachment handle more complex shapes. The drive system also helps to keep it going for anything up to four hours, meeting the demands of long-duration tasks.
To train the system’s motion control and navigation algorithms, says Professor Liu, “We started by creating a lifelike virtual world that provided a platform for the robot to learn walking, climbing and dodging. Through extensive trials and learning, it gained the ability to balance and navigate complex environments. Following this learning phase, we applied the algorithm to the actual robot and adjusted it to suit real world situations.”
The intelligent handler
The AI-powered dual-arm humanoid platform marks a significant step in robotics by effectively integrating task understanding with physical execution. This approach allows for greater versatility and coherence compared to traditional models, which often focus on isolated actions.
The dual-arm system is powered by what’s called a vision language model, a type of AI model that can understand and interpret both visual material and text, allowing it to understand complex spatial environments. It excels at deconstructing tasks in intricate scenarios and strategically planning subsequent subtasks. Integrated with robotic control algorithms, this enables diverse real-world operations, such as household chores.
A revolution in automation
The potential applications are vast, the most obvious being factories, warehouses, retail, delivery, and agriculture. It could even have future household applications – one day, you might find yourself buying an M1 or a dual-arm robot to do the housework for you.
The team is now working to further refine the system, focusing on automated battery replacement solutions, enhanced cognitive abilities, multi-robot collaboration, and more cost-effective designs.
Having been tested extensively in labs, outdoors, and in factories and logistics facilities, the next step, says Professor Liu, is commercialisation. The team has incubated a company, Lightyear Robotics Limited, which is in discussion about partnerships with leading companies in the logistics and industrial sectors.
