Abstract:

Real-time interception of fast-moving objects by a robotic arm in cluttered environments filled with static or dynamic obstacles is an extremely challenging task, requiring reaction times of mere milliseconds. Existing robotic planning algorithms struggle to perform multiple skills, such as catching a dynamic object and avoiding obstacles simultaneously. This paper proposes a unified framework for robotic path planning that embeds high-dimensional temporal information contained in the event stream to distinguish between safe and colliding trajectories, in a low-dimensional space that is represented by a pre-constructed 2D densely connected graph. Our approach leverages a fast graph-traversing strategy to generate motor commands that enable the robotic arm to effectively avoid approaching obstacles while simultaneously intercepting fastmoving objects. One of the most distinctive features of our methodology is that it enables object interception and obstacle avoidance to be performed within the same algorithmic framework based on deep manifold learning. By leveraging highly efficient diffusion-map-based variational autoencoding and Extended Kalman Filter (EKF), we demonstrate the effectiveness of our approach on an autonomous 7-DoF robotic arm using only onboard sensing and computation. Our robotic manipulator is capable of avoiding obstacles of varying sizes and shapes while successfully capturing a fast-moving soft ball thrown by hand at normal speed from different angles.

Motion Planning Step 1: Avoiding Static Obstacles to Reach a Static Target Location

Motion Planning Step 2: Intercepting a moving target by predicting future location through EKF

Motion Planning Step 3  :  Multiple Skill Learning by Manifold Data Graph Traversing : Intercepting A Flying Target While Avoiding Moving Obstacles

Simulation Platform – Robotics Toolbox for Python 

Supplementary Material