Autonomous Excavator System Works on Various Machine Sizes

July 12, 2021

An autonomous excavator system (AES) has been introduced by Baidu Research Robotics and Auto-Driving Lab (RAL) and the University of Maryland, College Park. The system can perform material loading tasks without any human intervention while offering performance closely equivalent to that of an experienced human operator, according to Baidu.

“This work presents an efficient, robust, and general autonomous system architecture that enables excavators of various sizes to perform material loading tasks in the real world autonomously,” said Dr. Liangjun Zhang, head of Baidu Research Robotics and Auto-Driving Lab, in a prepared statement.

According to Baidu, this is among the world’s first uncrewed excavation systems to have been deployed in real-world scenarios and continuously operating for over 24 hours. The researchers described their methodology in a research paper published on June 30, 2021, in Science Robotics.

AES uses accurate and real-time algorithms for perception, planning, and control alongside a new architecture to incorporate these capabilities for autonomous operation. Multiple sensors—including LiDAR, cameras, and proprioceptive sensors—are integrated for the perception module to perceive the 3D environment and identify target materials, along with advanced algorithms such as a dedusting neural network to generate clean images.

With this modular design, the AES architecture can be effectively utilized by excavators of all sizes—including 6.5- and 7.5-metric-ton compact excavators, 33.5-metric-ton standard excavators, and 49-metric-ton large excavators, the groups say. In the tweet below, the system is shown on an XCMG excavator at Bauma China.

Field testing the autonomous excavator

To evaluate the efficiency and robustness of AES, researchers teamed up with XCMG and other major manufacturers to deploy the system at a waste disposal site, a toxic and harmful real-world scenario where automation is in strong demand. The system was able to continuously operate for more than 24 hours without any human intervention, according to Baidu. AES has also been tested in winter weather conditions, where vaporization can pose a threat towards the sensing performance of LiDAR. The amount of materials excavated, in both wet and dry form, was 67.1 cubic meters per hour for a compact excavator, which is in line with the performance of a traditional human operator.

“AES performs consistently and reliably over a long time, while the performance of human operators can be uncertain,” said Dr. Zhang.

Researchers also set up 10 different scenarios at a closed testing field to see how the system performed in numerous real-world tasks. After testing a variety of large, medium-sized, and compact excavators, AES was ultimately proven to match the average efficiency of a human operator in terms of the amount of materials excavated per hour.

“This represents a key step moving towards deploying robots with long operating periods, even in uncontrolled indoor and outdoor environments,” said Dr. Dinesh Manocha, professor of computer science and electrical and computer engineering at the University of Maryland, College Park.

The research paper, “An autonomous excavator system for material loading tasks,” L. Zhang, J. Zhao, P. Long, L. Wang, L. Qian, F. Lu, X. Song, D. Manocha, was published in the journal Science Robotics on June 30, 2021.

Baidu Research RAL focuses on robotics and computer vision research, applying innovative research to autonomous driving, industry, and service robots. By combining expertise in robotics, computer vision, machine learning, simulation and systems, and studying holistic robotics solutions, RAL strives to be the first to bring research lab results, either internally or externally, to market.

Source: Baidu Research