PIPE: Process Informed Parameter Estimation, a learning based approach to task generalized system identification

TL;DR

This work addresses the problem of robot guided assembly tasks, by using a learning-based approach to identify contact model parameters for known and novel parts using a newly proposed neural network structure.

Abstract

We address the problem of robot guided assembly tasks, by using a learning-based approach to identify contact model parameters for known and novel parts. First, a Variational Autoencoder (VAE) is used to extract geometric features of assembly parts. Then, we combine the extracted features with physical knowledge to derive the parameters of a contact model using our newly proposed neural network structure. The measured force from real experiments is used to supervise the predicted forces, thus avoiding the need for ground truth model parameters. Although trained only on a small set of assembly parts, good contact model estimation for unknown objects were achieved. Our main contribution is the network structure that allows us to estimate contact models of assembly tasks depending on the geometry of the part to be joined. Where current system identification processes have to record new data for a new assembly process, our method only requires the 3D model of the assembly part. We evaluate our method by estimating contact models for robot-guided assembly tasks of pin connectors as well as electronic plugs and compare the results with real experiments.

Figures (9)

• Figure 1: Depiction of the variables describing the robot guided assembly task. The first picture shows a dowel pin and the second picture a DSUB37 plug. $\Omega_i$ is the object at assembly task $i$, $F_z$ the resulting force during assembly and $p_z$ the positon during contact.
• Figure 2: Setup of the robot guided assembly task, consisting of (1) force-torque-sensor FTN-AXIA80, (2) gripper SCHUNK EGH 80-IOL-N, (3) task board and (4) UR5e robotic arm.
• Figure 3: Overview of the approach. $P_i$ corresponds to process information of different assembly tasks $i$. $NN(\cdot)$ is the neural network mapping the process information to model parameters $\underline{\Theta}_i$. $u$ and $x$ describe the system input and state variable respectively.
• Figure 4: Selection of training data examples used in different task domains. In each column, the mesh of the object and the measured force $F_z$ in blue and position profile $p_z$ in black during assembly is shown. The shaded area depicts the variance and the solid line the arithmetic mean of the $10$ different assembly tasks.
• Figure 5: Network architecture of the used VAE model. First five blocks indicate 1D convolution with batch normalization and ReLU activation, other blocks indicate fully connected layer. $\odot$ means element-wise product. $\Omega_i$ is the input geometry and $\Omega_i'$ the reconstructed input. $l$ stands for the latent-space dimension and $n$ for the number of points in the point cloud.