ForceEstimation

Neural Network based external force estimation on dVRK.

Network implementation for real time application, based on the architecture defined in the paper titled "Neural Network based Inverse Dynamics Identification and External Force Estimation on the da Vinci Research Kit" by Nural Yilmaz et al.

Network architecture

Layer 1

Input Size: 12, one set of Position and Velocity measurement from each of the 6 joints.
Output Size: 100
Activation: tanh

Layer 2

Input Size: 100
Output Size: 1

6 such networks are used. Each network is trained to provide a torque estimate of one the joints. This torque is then used for the Force estimate.

Commands

For training the network(uses given rosbag file)

python3 main.py train <Path to rosbag file> <Enter number of epochs here>

For running inference using a saved model(currently uses the same hardcoded array as used for training)

python3 main.py inference