Researches in recent years have been explored the use of robotic arms or dexterous hands to solve everyday tasks. Simple task such as grasping or basic manipulation have been tackled successfully, while complex tasks involving multiple steps and strategic movements have so far proved harder to address. Researches team from Chinese University of Hong Kong and Tencent AI Lab has recently developed a deep learning-based approach to solve a Rubik's Cube using a multi-fingered dexterous hand. The team developed a hierarchical deep reinforcement learning model to slove the Rubik's Cube puzzle using a dexterous hand the that essentially separates the task into a planning and a manipulation stage. The researchers applied this approach to a five-fingered dexterous hand called the Shadow Hand. Tackling a Rubik's Cube using a robotic hand are two-fold, first controlling a robotic hand is very difficult since it has a high degree of freedom and second solving a Rubik's Cube requires a long motion sequence. The hierarchical model has two key components one for planning and one for manipulation.The planning component initially identifies the optimal mode sequence for solving the puzzle and then, the manipulation controller controls the dexterous hand's fingers to execute these steps. The model was trained and evaluated by the Researchers using a high-fidelity simulator and tested in a series of experiments in which the dexterous hand was fed with 1400 randomly scrambled Rubik's Cubes and it achieved an average success rate of 90.3 percent.
0 Comments