Learning to teleoperate an upper-limb assistive humanoid robot for bimanual daily-living tasks

Objective. Bimanual humanoid platforms for home assistance are nowadays available, both as academic prototypes and commercially. Although they are usually thought of as daily helpers for non-disabled users, their ability to move around, together with their dexterity, makes them ideal assistive devices for upper-limb disabled persons, too. Indeed, teleoperating a bimanual robotic platform via muscle activation could revolutionize the way stroke survivors, amputees and patients with spinal injuries solve their daily home chores. Moreover, with respect to direct prosthetic control, teleoperation has the advantage of freeing the user from the burden of the prosthesis itself, overpassing several limitations regarding size, weight, or integration, and thus enables a much higher level of functionality. Approach. In this study, nine participants, two of whom suffer from severe upper-limb disabilities, teleoperated a humanoid assistive platform, performing complex bimanual tasks requiring high precision and bilateral arm/hand coordination, simulating home/office chores. A wearable body posture tracker was used for position control of the robotic torso and arms, while interactive machine learning applied to electromyography of the forearms helped the robot to build an increasingly accurate model of the participant’s intent over time. Main results. All participants, irrespective of their disability, were uniformly able to perform the demanded tasks. Completion times, subjective evaluation scores, as well as energy- and time- efficiency show improvement over time on short and long term. Significance. This is the first time a hybrid setup, involving myoeletric and inertial measurements, is used by disabled people to teleoperate a bimanual humanoid robot. The proposed setup, taking advantage of interactive machine learning, is simple, non-invasive, and offers a new assistive solution for disabled people in their home environment. Additionnally, it has the potential of being used in several other applications in which fine humanoid robot control is required.

Human Factors in Interactive Machine Learning: A Cybersecurity Case Study

Cybersecurity is emerging as a major issue for many organizations and countries. Machine learning has been used to recognize threats, but it is difficult to predict future threats based on past events, since malicious attackers are constantly finding ways to circumvent defences and the algorithms that they rely on. Interactive Machine learning (iML) has been developed as a way to combine human and algorithmic expertise in a variety of domains and we are currently applying it to cybersecurity. In this application of iML, implicit knowledge about human behaviour, and about the changing nature of threats, can supplement the explicit knowledge encoded in algorithms to create more effective defences against cyber-attacks. In this paper we present the example problem of data exfiltration where insiders, or outsiders masquerading as insiders, who copy and transfer data maliciously, against the interests of an organization. We will review human factors issues associated with the development of iML solutions for data exfiltration. We also present a case study involving development of an iML solution for a large financial services company. In this case study we review work carried out on developing visualization dashboards and discussing prospects for further iML integration. Our goal in writing this paper is to motivate future researchers to consider the role of the human more fully in ML, not only in the data exfiltration and cybersecurity domain but also in a range of other applications where human expertise is important and needs to combine with ML prediction to solve challenging problems.

Interactive Machine Learning-Based Multi-Label Segmentation of Solid Tumors and Organs

We seek the development and evaluation of a fast, accurate, and consistent method for general-purpose segmentation, based on interactive machine learning (IML). To validate our method, we identified retrospective cohorts of 20 brain, 50 breast, and 50 lung cancer patients, as well as 20 spleen scans, with corresponding ground truth annotations. Utilizing very brief user training annotations and the adaptive geodesic distance transform, an ensemble of SVMs is trained, providing a patient-specific model applied to the whole image. Two experts segmented each cohort twice with our method and twice manually. The IML method was faster than manual annotation by 53.1% on average. We found significant (p < 0.001) overlap difference for spleen (DiceIML/DiceManual = 0.91/0.87), breast tumors (DiceIML/DiceManual = 0.84/0.82), and lung nodules (DiceIML/DiceManual = 0.78/0.83). For intra-rater consistency, a significant (p = 0.003) difference was found for spleen (DiceIML/DiceManual = 0.91/0.89). For inter-rater consistency, significant (p < 0.045) differences were found for spleen (DiceIML/DiceManual = 0.91/0.87), breast (DiceIML/DiceManual = 0.86/0.81), lung (DiceIML/DiceManual = 0.85/0.89), the non-enhancing (DiceIML/DiceManual = 0.79/0.67) and the enhancing (DiceIML/DiceManual = 0.79/0.84) brain tumor sub-regions, which, in aggregation, favored our method. Quantitative evaluation for speed, spatial overlap, and consistency, reveals the benefits of our proposed method when compared with manual annotation, for several clinically relevant problems. We publicly release our implementation through CaPTk (Cancer Imaging Phenomics Toolkit) and as an MITK plugin.

Interactive biomedical segmentation tool powered by deep learning and ImJoy

Deep learning-based methods play an increasingly important role in bioimage analysis. User-friendly tools are crucial for increasing the adoption of deep learning models and efforts have been made to support them in existing image analysis platforms. Due to hardware and software complexities, many of them have been struggling to support re-training and fine-tuning of models which is essential to avoid overfitting and hallucination issues when working with limited training data. Meanwhile, interactive machine learning provides an efficient way to train models on limited training data. It works by gradually adding new annotations by correcting the model predictions while the model is training in the background. In this work, we developed an ImJoy plugin for interactive training and an annotation tool for image segmentation. With a small example dataset obtained from the Human Protein Atlas, we demonstrate that CellPose-based segmentation models can be trained interactively from scratch within 10-40 minutes, which is at least 6x faster than the conventional annotation workflow and less labor intensive. We envision that the developed tool can make deep learning segmentation methods incrementally adoptable for new users and be used in a wide range of applications for biomedical image segmentation.