University of Twente Student Theses

Login

Robotic bin-picking pipeline for chicken fillets with deep learning-based instance segmentation using synthetic data

Jonker, L.M. (2023) Robotic bin-picking pipeline for chicken fillets with deep learning-based instance segmentation using synthetic data.

[img] PDF
11MB
Abstract:Abstract—In the food processing industry, automation is getting more common for purposes such as increased food quality and compensation of worker shortages. An automation task such as gripping is challenging due to deformation of the objects. Additionally, in order to manipulate food in a specific manner, it is important to know the location and orientation of the food object. Due to these deformation and location problems, automation of tasks such as bin-picking food objects is a difficult challenge. Existing automated bin-picking methods for food objects lack environmental awareness and the ability to recreate the scene and objects in 3D in order to manipulate them and avoid collisions. In this research, we present a robotic pipeline for bin-picking of chicken fillets. The individual chicken fillets are detected using instance segmentation via deep learning. The instance segmentation model is trained on synthetically rendered images of chicken fillets. Current methods for synthetic data generation only use rigid body simulations, whereas we also simulate the deforming physics on manually created 3D models. Additionally, the path planning is based on a 3D reconstructed environment, using depth data from an RGB-D camera. The tests on the instance segmentation model with real data yield a bbox and mask AP@50:5:95 of 0.74 and 0.68, respectively. Finetuning the model with a small real dataset increases the APs to 0.82 and 0.78, respectively. Tests on the full pipeline show a planning success rate of 0.73, and an execution success rate of 0.81. We show that automation of bin-picking for chicken fillets using synthetic data is a realistic prospect. A supplementary video showcasing the pipeline can be found on https://tinyurl.com/5n8v8uf7.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Programme:Systems and Control MSc (60359)
Link to this item:https://purl.utwente.nl/essays/94947
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page