In context
In early 2025, robotic grasping remained a bottleneck for automation in unstructured environments such as logistics and warehousing. While data-driven methods had improved grasp detection, many failed to ensure stability during transport, especially for irregularly shaped objects that could shake or slip from grippers. This paper from South China University of Technology addressed that gap by combining shape analysis with probabilistic reasoning.
What was reported
The proposed technique first analyzes the point cloud of an object to assess its dispersivity and flatness, classifying it by irregularity degree (ratio of surface area to volume). Candidate grasp poses are generated using an antipodal point algorithm that filters contact points based on shape features, such as clustering and principal component analysis, to better align with the object's center of mass.
To evaluate grasp stability during transport, the authors introduced a probabilistic reasoning method. They simulated grasping, rotation, and translation tasks with domain randomization, counting successful trials. A conditional expectation formula quantified the likelihood of stable transport given a successful grasp. A PointNet discriminator was trained on this simulation data to score and rank candidate poses.
Experiments showed an average grasp success rate of 89.2%, a 2.6% improvement over a baseline, and transport stability of 84.2%, a 22.7% improvement. The method also enabled intelligent grasping in multi-object stacking scenarios, maintaining stability and logical grasp ordering.
Why it mattered
This work highlighted the importance of considering post-grasp transport stability, not just initial grasp success, in robotic manipulation. By integrating shape analysis with probabilistic evaluation, it offered a practical approach for handling irregular objects in logistics and other industrial settings, potentially improving the reliability of automated picking and placing.
“The proposed method enables intelligent grasping of objects in multi-object stacking scenarios, ensuring stability during the grasping and transporting process, and establishing a logical sequence for grasping.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2025-01-15 · “基于形状分析和概率推理的机器人抓取技术”