Autonomous Dynamic Grasping with a Robotic Arm: Real-Time Motion Prediction and Adaptive Control

Dynamic grasping of moving objects represents one of the most challenging and actively researched areas in robotic manipulation. Unlike static grasping, this task requires the integration of perception, prediction, and control in real-time. Typical approaches involve tracking the target’s position in real-time using vision-based systems, estimating its velocity, and generating tailored Cartesian-space trajectories to ensure the robot’s end-effector coincides with the object's position at the precise time required for successful grasping. The presented work addresses the development, optimization, and hardware implementation of a novel control algorithm for dynamic grasping. The research utilizes the ROS 2 framework alongside the WidowX 250 S robotic arm, which features six degrees of freedom. This configuration allows for autonomous grasping of objects in motion while dynamically adapting to alterations in their trajectory, which are not necessarily restricted in space. To achieve this objective, a motion capture system continuously monitors the target's position relative to the manipulator, streaming real-time data to the robot's controller, which estimates the relative velocity. Based on these estimates and predefined constraints, the controller intelligently determines the optimal timing to initiate the grasping sequence, predicting the target's trajectory and evaluating its transit in the dynamically computed reachable workspace of the robotic arm, determined at runtime using an efficient analytic inverse kinematics function tailored to the robot. Additionally, an advanced adaptive trajectory planner generates detailed trajectories, including position, orientation, velocity, and acceleration for the robot's end-effector, in under a millisecond. These trajectories are then relayed to the motor controllers, ensuring precise and coordinated motion. Finally, experimental validation confirms that the developed adaptive dynamic grasping algorithm performs as expected, demonstrating high repeatability and accuracy, effectively showcasing its robustness in real-world scenarios.