A Hybrid Neuromorphic Robotic-Arm Lemniscate Trajectories Dataset for Learning-Based Applications

In this work, an all-in-one neuromorphic motors controller system with reduced latency and power consumption for a robotic arm is used to record a dataset of trajectories for learning applications. Unlike human-created encoding systems, biological neural ones use neurons and spikes to process the information and make weighted decisions based on a continuous learning process. A central pattern generator is a biological structure that orchestrates the activity of motor neurons, causing muscle fibers to elongate and contract through signals generated by spikes. This process allows for the replication of learned movement patterns or trajectories. The Event-Driven Scorbot platform (ED-Scorbot) consists of a six Degrees of Freedom (DoF) robotic arm, whose controller implements a Spiking Proportional-Integrative-Derivative (SPID) algorithm, mimicking biological systems. In this paper, the ED-Scorbot is used to record a series of lemniscate trajectories drawn by the end-effector in the Cartesian space. These trajectories have been produced with different heights, widths, foci distances and orientation angles within the robot's working area. Different weights at the end-effector are used to enrich the dataset. MP4 videos, internal spiking activity, joint sensors output and target trajectory are included in this hybrid dataset. For its validation, the spiking activity has been used to train and test a recurrent Spiking Neural Networks (SNN) to classify the weight attached to the gripper with an accuracy of 83.3% on test samples.