Finger-position optimization by using caging qualities

Caging aims at constraining objects like a bird cage. It does not need force closure and contacts, and therefore is robust to perception noises. This paper considers these merits of caging and proposes an efficient approach to find the optimized finger positions for robust grasping under perception noises by measuring the quality of caging. It initially uses two quality models (the quality model of translational constraint and the quality model of rotational constraint) to combinatorically represent the quality of caging and finds the optimized finger positions (finger positions that have large robustness to perception noises) by maximizing the margins to caging breaking. Meanwhile, it employs Genetic Algorithm (GA) to accelerate the combinatorial optimization of the two quality models. The encoding rules, the initial procedure, and the stopping criterion of the GA are designed carefully towards the quality models to improve its computational efficiency. Experiments with various object shapes show that with the help of multi-model optimization and GA acceleration, the proposed approach is not only robustness to perception noises but also efficient in computation.