Low-complexity Reinforcement Learning Decoders for Autonomous, Scalable, Neuromorphic intra-cortical Brain Machine Interfaces

General Description. TThis dataset comprises recordings from four BMI (Brain-Machine Interface) experiments conducted on two adult macaques. Three of the experiments involved joystick-controlled tasks, while the fourth was a center-out reaching task. In the center-out task, the macaque was trained to maneuver a joystick-controlled cursor from a central position on a computer screen to one of eight square-shaped target locations. The macaques were able to use a wireless integrated system to control a robotic platform (on which they were seated) enabling independent mobility driven by neuronal activity in their motor cortices. Neural activity was recorded from populations of single neurons via multiple electrode arrays implanted in the arm region of the primary motor cortex. A general overview is provided below:

1. A titanium head post (Crist Instruments, MD, USA) was surgically affixed before implanting the microelectrode arrays. In NHP-A, four microelectrode arrays with 16 electrodes each were implanted, while NHP-B was implanted with one array containing 100 electrodes in the hand/arm region of the left primary motor cortex.
2. Spike signals were recorded using an in-house 100-channel wireless neural recording system, sampled at 13 kHz. The wide-band signals were band-pass filtered between 300 and 3000 Hz to eliminate low-frequency components. Spike detection thresholds were determined using the formula: Thr = 5σ, where σ = median(|x| / 0.6745), x is the filtered signal, and σ estimates the standard deviation of background noise.

In Experiments 1, 2, and 3, the behavioral task involved controlling the motion of a robotic wheelchair using a three-directional, spring-loaded joystick. These tasks included: a) turning 90° right, b) moving forward by 2 meters, c) turning 90° left, and d) remaining stationary for 5 seconds (stop task). The success rate varied across experiments. Experiment 4 also involved joystick control, but followed a classical center-out reaching paradigm. 

Dataset Description. The dataset is organized into folders (labeled as experiment 1, 2, 3, and 4) containing data from both NHP-A and NHP-B. Each folder contains data from separate dates labeled as YYMMDD (at the end of the filename). For experiment 1, data from the following dates are present: 15-10-08, 15-10-12, 15-10-19, 15-10-26, 15-11-02, 15-11-16, 15-11-23, and 15-12-10. For experiment 2, following dates are: 18-02-20, 18-03-06, 18-03-08, 18-03-20, 18-03-26, 18-04-13, 18-04-16, 18-04-23. For experiment 3: 14-08-14, 14-08-18, 14-08-20, 15-10-14. For experiment 4: 18-12-03, 18-12-13, 19-01-07, 19-02-20.  File naming conventions across all experiments are as follows

1. targTest: This corresponds to the direction of the joystick recorded for each trial. (decoded using the decoder)
2. targTrain: Ground truth label, corresponding to the actual direction of the joystick.
3. testSet: Number of spike counts from each channel (used for testing corresponding to all the sessions)
4. trainSet: Number of spike counts from each channel (used for calibration, mostly)       

Additional Information. This dataset is a simplified and curated version designed to reproduce the results presented in the associated paper. Note that Experiments 1 and 3 have partial datasets already publicly available at: https://osf.io/dce96/. However, those versions are raw and can be processed using the variable descriptions below to extract spike counts with a specified bin width. Each file includes the following fields:

1.  joystick_adfreq: The frequency of operation of the joystick.
2. X_Voltage: The voltage reading corresponding to the x-coordinate (while joystick operation).
3. Y_Voltage: The voltage reading corresponding to the y-coordinate (while joystick operation).
4. Spike_data(Channel Number): The Channel Number corresponding to which the neuronal data is recorded.
5. Spike_data(Cluster): Descripting the cluster on which the channels are placed.
6. Spike_data(Spike Times): The timestamp corresponding to the detection of a spike.
7. Spike_data(Spike Number): The total number of spikes calculated for a particular trial from a particular channel.
8. Spike_data(Mean Spike Waveform): The mean neuronal data (for that trial from a particular channel) describing a spike.
9. Spike_data(Spike Amplitude): The mean spike amplitude of that particular channel.
10. IMETrainingData(SentSignals): The truth labels corresponding to a particular trial.
11. IMETrainingData(Timestamps): Time stamps corresponding to each sent signal (data).
12. IMETrainingData(ReasonFail): String data; Reason if the trial failed.
13. IMETrainingData(TrialOutcomes): Trial results corresponding to successful or unsuccessful!
14. IMETrainingData(StartTime): corresponding to the beginning of each trial.
15. IMETrainingData(EndTime): corresponding to the end of each trial.

Possible use cases. This dataset is well-suited for designing, training, and evaluating iBMI decoders. It provides a valuable resource for researchers aiming to model sensorimotor cortical spiking, benchmark iBMI decoders under consistent conditions, or explore neuromorphic and reinforcement learning-based approaches to decoder design.

Contact Information. We would be delighted to hear from you if you find this dataset useful—especially if it contributes to a publication. Contact: A. Basu <arinbasu@cityu.edu.hk>; A. Ghosh <aghosh14@illinois.edu>.

Citation. A. Ghosh, S. Shaikh, B. Zhou, P. S. V. Sun, C. Libedinsky, R. So, A. Basu, "Low-complexity Reinforcement Learning Decoders for Autonomous, Scalable, Neuromorphic intra-cortical Brain Machine Interfaces," Neuroelectronics 2025(2):0006, https://doi.org/10.55092/neuroelectronics20250006