Tech

EfficientGrasp: A unified data-efficient learning method for multi-fingered robotic hands


Capture multiple objects is one of the most important abilities of the robot hand. However, current techniques for this are often not directly applicable to new grippers of different shapes or configurations.

A humanoid robot.

A humanoid robot. Image credits: PiqselCC0 . public domain

A recent paper on arXiv.org presents an improved data-driven capture aggregation and gripper control approach. It effectively generalizes to novel general-purpose grippers regardless of geometry and kinematics and does not use hand model specifications.

This method creates contacts based on the fingertip workspace of a clamp. A model-free reinforcement learning method is applied to compute the inverse kinematics of the clamp; therefore, the method can be extended to new grippers that do not require a kinematic model.

Compared to competing methods, the proposed approach is more data-efficient, more accurate, and can produce more reliable information.

Automatically capturing novel objects previously invisible to robots is a constant challenge in robot manipulation. Over the past decades, many approaches have been proposed to solve this problem for specific robot hands. The UniGrasp framework, introduced recently, has the ability to generalize to different types of robotic grippers; however, this method does not work on grippers with closed loop constraints and is not data efficient when applied to multigrasp configurations. In this paper, we present EfficientGrasp, a generalized method of capturing and controlling the gripper independent of the specifications of the gripper model. EfficientGrasp uses the gripper workspace feature rather than the UniGrasp gripper property inputs. This reduces memory usage during training by 81.7% and makes it possible to generalize to a wider range of grippers, such as clampers with closed loop constraints. EfficientGrasp’s performance is evaluated by performing object capture experiments both in simulation and in the real world; The results show that the proposed method is also superior to UniGrasp when considering only clampers without closed-loop constraints. In these cases, EfficientGrasp showed a 9.85% higher accuracy in creating touchpoints and a 3.10% higher capture success rate in the simulation. Real world experiments were conducted with a clamp with closed loop constraints, which UniGrasp failed to handle while EfficientGrasp achieved a success rate of 83.3%. The main causes of capture failure of the proposed method are analyzed, highlighting ways to enhance capture performance.

Research articles: Li, K., Baron, N., Zhang, X. and Rojas, N., “EfficientGrasp: A unified data-efficient capture method for multi-fingered robotic hands”, 2022. Link: https://arxiv.org/abs/2206.15159






Source link

news7g

News7g: Update the world's latest breaking news online of the day, breaking news, politics, society today, international mainstream news .Updated news 24/7: Entertainment, Sports...at the World everyday world. Hot news, images, video clips that are updated quickly and reliably

Related Articles

Back to top button