It is my great pleasure to announce that I was awarded the Excellent New Joiner award for the year 2023 at Huawei Hong Kong.

It was a honor for me to receive such an award from the company after my first year of joining it. I hope and I wish I will be able to keep contributing and give more for my personal and for the company’s success.

It is my great pleasure to announce that our latets paper has been publishe on IEEE Transaction on Instrumentation and Measurements (TIM).

The work is about image classification for defect detection in LCD manufacturing, specifically open-set classification.

Qualty assessment is of uttermost importance in industries like LCD manufacturing, in order to ensure products meet customer’s requirements and saving time and costs in the production.

Machine Learning for computer vision has had great advancements over the years and now there exist powerful models that are able to detect and segments objects in images. These models are, however, trained on large-scale open-source datasets, and do not perform well on smaller, industry-specific ones.

Additionally, these detectors are generally trained in closed-world assumptions, namely it is assumed that the classes used during training will be the same used in testing (during production).

However, this assumption is not always met, especially in the manufacturing industry, where new unknown defects may appear and be completely different from the known ones.

The goal of this work is to evaluate how state-of-the-art approaches for open-set classification perform on industry-specific datasets, and we additionally propose a novel model which is capable of properly classifying both the known and unknown classes, named Error Cluster Classifier.

In this work we compare:

• Baseline Thresholded Softmax Classifier: which is a standard classifier model which identifies as unknonw classes those whose predicted probability is below a certain threshold.
• OpenMax Classifier: which is similar to a standard classifier but with an additional head that models the probability distribution as a Weibull distribution.
• G-ODIN: a model for identifying out-of-distribution data. G-ODIN takes an image classifier and splits the output into two functions, one to approximate the probability of the input of belonging to the in-distribution domain, and the other one predicting the joint probability of the input of belonging both to the in-distribution domain and to a known class.
• Open Long-Tailed recognition (OLTR): which consists of a model to cluster features extracted from a backbone network. The extracted features are then fed into a Meta-Embedding classifier which predicts the final class probability. During training, the loss function tries to cluster the known class features appropriately.
• Error Cluster Classifier: it is our proposed model that takes inspiration from OLTR. The main difference is in the classifier itself, which simply considers the distance of the backbone’s extracted features from learned class centroids to estimate the probability of the input of belonging to known classes. During training, the loss function tries to cluster the extracted features, such that features of the same class belong to the same cluster and are pushed away from the non-class clusters.

In this work we also propose a novel loss for clustering the features, named Strongly Repulsiv (SR) loss. Differently from the standard Low Repulsive (LR) clustering loss, the SR loss tries to cluster extracted features by forcing a given distribution.

We tested the approaches on two different LCD datasets.

In conclusion, our work shows the difficulties of state-of-the-art approaches in dealing with unseen classes. From our experiments, it is evident that feature clustering approaches like OLTR and CE cClassifier perform generally better.

Overall, the proposed CE classifier is the one that perfromrs the best on both datasets, given its capability of predicting the input’s class by estimating the distance of the extracted features from the trained known class centroids.

It is my great pleasure to announce that I was selected as one of the finalist of the Amazon Award 2022 organized by Imperial College Robotics Forum.

I presented the research achievents of my PhD on the novel approahces developed for robot model learning, control, and design optmization.

Accepted for publication in IEEE Transaction in Automation Science and Engineering (T-ASE), 2022

Description:

Machine learning and Artificial Neural Networks (ANN) can be powerful tools to model complex systems like robots.
In fileds like surgery, robots are very complex, and analytical models are hard to compute. Having an accurate model is, however, fundamental in order to ensure safety and precise motion, espeially becuase sensors can hardly be used in such scenarios.

One major limitiation of ANN, is that they are deterministic, meaning that they only provide a “punctual” prediction, given an input.
However, the models and the data have uncertainties themselves, and an estimation of these uncertainties could be beneficial to have an estimation of the confidence level of the model.

Probabilistic Neural Networks can solve this problem. Approaches like Bayesian Neural Networks (BNN) and Evidential Neural Networks (EvNN) allow having not only the model’s prediction, but also the setimation of the model’s uncertainty (epistemic undertainty) and data uncertainty (aleatoric uncertainty).

In this work we propose the use of BNN and EvNN to model the forward kinematics of a surgical robot.
The surgical instrument is mounted on top of a serial-link manipulator in a macro-micro manipulator setup.

For the control of the overall system, to perform a surgical task, we employ Hierarchical Quadratic Programming (HQP) in order to exploit the hyper-redundancy of the system and impose two prioritized tasks:
1) ensuring fulcrum effect at the insertion point location, also known as Remote Centr of Motion (RCM) constraint;
2) tracking a given path, to simulate a tumor resection task.

Additioanlly, we include a lower level task solved by means of a Sequential Qudaratic Programming (SQP) probelm in order to minimize the model’s uncertainty during operation and ensure the robot perfroms the task with the highest confidence possible.

Our simulation and real world tests show the importance of minimizing model’s unceratinties in order to guarantee safety and improve task execution.

In fact, without the uncertainty minimization, the robot doesn’t manage to accurately track the desired paths; much higher accuracy is instead achieved with our proposed approach.

It is my great pleasure to announce that I presented two of my PhD works at IROS 2022, one of the top international robotics conferences, held in Kyoto.

The two works are those described in :

• Augmented Neural Network for Full Robot Kinematic Modelling in SE(3), published in IEEE Robotics and Automation Letters (RAL), about including differential relationships during model learning with Artificial Neural Networks for robot kinematics. You can check the brief summary here and the full paper here.
• Optimization of Surgical Robotic Instrument Mounting in a Macro–Micro Manipulator Setup for Improving Task Execution, published in IEEE Transaction on Robotics (T-RO), about an optimization framework for identifying the optimal mounting connection between a serial-link robotic arm and a surgical robotic tool. You can check the summary here and the paper here.

I am here sharing the videos of my presentations during the conference.

Accepted for publication in IEEE Robotics and Automation Letters (RAL), 2022

Description:

Surgical robots are very complex systems, due to the requirements for miniaturization and high dexterity. This leads to highly articulated and complex designs to increase the system’s felxibility and dexterity, while the miniaturization requirement leads to the choice of non-traditional actuation systems like tendons.

Differently from traditional serial-link manipulators, generally used in the manufacturing industry, tendon-driven robots do not have motors directly on the joints, but they are placed on the robot’s base and connected to the joints by means of wires (the tendons).

This setup causes a lot of complexities in the system due to high nonlinearities due to tendon friction, slack, tendon elongation, that, in turn, cause effects like backlash and hysteresis.

Hystereis is phenomenon that causes a system not to behave in the same way when moving in a direction or in the opposite one. Backlash occurs when a change in motion direction occurs and, despite commanding non-null velocities, the system lies in a deadzone and does not move. In tendon-driven system this is typically due to the tendon elongation and possible slack when motors change their direction of motion.

Machine learning (ML) techniques have been very efficient in building complex robot models, especially for surgical robots that are very hard to model. These approaches, however, do not consider andy a priori knowledge of backlash and hysteresis and if and end-to-end black box model is used, the model should be able to inherently learn a compensation strategy. This makes the application of ML moelling tedious in surgical robots.

In this work we propose a backlash compensation technique and use Artificial Neural Networks to model the complexities of a tendon-driven surgical robot from the compensated values.

The a priori backlash compensation is used to reduce the nonlinearities in the system and make it easier to learn a more accurate and simpler robot model.

With our backlash compensation approach we could reduce the occurrence of deadzones by 89%, thus limiting the stationary motion of the robot.

We then employ two different approaches to model our robot (a traditional ANN and our proposed AugNet, which incorporates differential relationships during training) and performed path tracking tests with and without the backlash compensation.

Our results show the importance of compansating for high nonlinearities before using ML approaches and that our compensation approach helps improving system’s tracking accuracy.