Join our full-day workshop on Autonomy in Robotic Surgery at ICRA 2024! Led by experts Paolo Fiorini and Riccardo Muradore, this workshop will dissect the regulatory challenges posed by the AI Act and delve into the integration of autonomous features into surgical robots.

With a lineup of renowned speakers and interactive sessions, anticipate a day of insightful discussions on the state of the art, technical advancements, and the roadmap for the future of autonomous robotic surgery.

ABSTRACTS

Fundamentals of Autonomy, Prof. Paolo Fiorini (retired) University of Verona (Italy)

Robotic surgery, or better robot-assisted surgery, is based on the teleoperation paradigm, in which one single operator, the surgeon, commands one single robotic device. The robot duplicates precisely the surgeon’s movements, modulo some motion scaling and filtering. However, this paradigm is very limiting since it requires the space and time co-existence of the robot and its human operator. Some level of independence by the robotic device will have profound implications, e.g. free surgical personnel, make more effective use of surgeons’ time, support tele-surgery, and potentially improve safety and efficiency. But the path to robotic autonomy in surgery is long and difficult. In this talk I will summarize what are the basic definitions and properties of autonomy, describe the work done at the University of Verona to develop technologies relevant to robotic autonomy and briefly discuss the risks and myths about the use of AI-based algorithms in supporting machine cognition.

Imaged-guided automation in human-robot collaborative surgery , Prof. Yunhui LIU, The Chinese University of Hong Kong 

The increasing robotic and AI techniques are expected to revolutionize modern surgery paradigms by endowing surgeons with enhanced operational capabilities, but AI-enabled surgical robotics for assisting surgeons to perform solo-laparoscopic surgery has rarely been developed nor pre-clinical validated. This is due to mixed challenges posed by the insufficient scene understanding of the complex surgical scenarios to enable intelligent control for surgical robots, lacking appropriate cooperation approaches between surgeons and robots to ensure efficiency and safety, and the intricacies involved in deploying clinical trials. This talk presents our recent efforts and outcomes on the image-guided surgical automation and human-robot collaboration system that offers next-level autonomy assistance to surgeons and enables them to solely perform a complete surgery. We first present the automatic laparoscope control method for providing an optimal filed-of-view during intraoperative maneuvers, which employs a novel learning-based framework for real-time detection and tracking of multiple surgical instruments in the view, facilitating precise and robust interaction in dynamic and complex surgical scenarios. We also introduce a novel graph-user-interface (GUI)-based semi-automatic control scheme that operates in a human-in-the-loop fashion, incorporating the primary surgeon’s high-level decisions to control the robot during surgery. This approach enhances the operating efficiency without the need of external sensing devices and accomplishes the solo-surgery in complex clinical setups. Furthermore, we present the application of the above surgical automation and the human-robot collaboration strategy in our developed one-surgeon-four-arms system, which integrates two surgical robots devised for laparoscope control and uterine manipulation in laparoscopic hysterectomy. It enables a single surgeon to efficiently perform the four-arm surgical tasks with enhanced coordination and has been validated in ex-vivo and in-vivo cadaver trails, showing the potential to revolutionize the current clinical paradigm and address the medical resource disparities issues.

Fundamental of Medical Regulations, Dr. Gernot Kronreif, ACMIT (Austria)

The main goal for autonomous medical robots is to achieve a higher level of  independence from user interaction in all four cognitive functions , i.e. in plan generation, plan selection, plan execution, and in monitoring. So far, higher autonomy levels have been reached primarily for plan execution, e.g. in external radiation therapy. Most work regarding autonomous robot setups, however, is still in the research phase and the translation into clinically available systems is still a long time coming. In this presentation, the focus is not so much on the possible technical and ethical hurdles for the realisation of autonomous medical robots, but on the regulatory requirements that need to be considered for translation of such systems to higher TRL levels.

Motivations of autonomy in robotic surgery, Prof. Alberto Arezzo,   University of Torino (Italy)

Robotic surgery is the fastest growing field of service robotics, thanks to strong acceptance by users, both surgeons and patients, and the benefits this technology brings to a range of procedures. To date, robots are sophisticated tools in the hands of surgeons and only provide tissue telemanipulation support, without any automatic support during the execution of a procedure. We will describe some of the steps needed to reach a new level of robotic surgery, i.e. adding (semi-)autonomous functions that will simplify surgical procedures, reduce cognitive and physical burden and increase patient safety. We will address two of the main challenges to autonomy, namely the availability of reliable research hardware, and will provide an overview of technologies now available or near-available in the clinical environment for different applications that would benefit from some degree of autonomy. Some of the applications of robotic autonomy will be described. We will point out some of the technical and non-technical challenges to robotic autonomy in a surgical context. In conclusion we will briefly address the ethical questions raised by the increasing levels of autonomy of surgical robots. These ethical issues are explored primarily with reference to cutting-edge case studies and upcoming advances in minimally invasive surgery (MIS) and microsurgery. In both areas, the surgical working space is limited and the precision required is high. For this reason, increasing levels of robotic autonomy can make a significant difference in these areas, and ethically justified sharing of control between humans and robots must. In particular, from the perspective of responsibility and capability, appropriate policies are proposed for the Significant Human Control (MHC) of increasingly autonomous surgical robots. It is highlighted how the MHC should be modulated according to the various levels of autonomy of MIS and microsurgery robots. Furthermore, newer MHC distinctions are introduced to handle contextual conditions regarding e.g. soft or rigid anatomical environments.

Robotic US platforms for guiding medical interventions, Prof. Arianna Mensiassi, Scuola Superiore San Anna Pisa (Italy)

This talk will describe medical robots (at different scales) which make use of US imaging for tracking and guidance of the intervention. Different case-studies will be considered, ranging from US-guided focused US surgery, to US guided catheter interventions, down to miniature US-tracked shuttles for targeted therapy. In all case studies, a special attention will be addressed to the role of the operator in the process.

Fundamental of Teleoperation, Prof. Riccardo Muradore, University of Verona (Italy)

Teleoperation is a well-known technique used in modern robot-assisted minimally invasive surgery. With the advent of autonomous or semi-autonomous systems, its role is expected to be increasingly less important. Unfortunately, unforeseen events are always possible and surgeons must be able to take control of the system in the event of a malfunction or critical patient situations that the “intelligence” of the system is not able to manage. Teleoperation will therefore come back into play to make the transition smooth and safe also according to the legislation which always requires the (possibly) active supervision of human beings.

Context awareness in robotic surgery, Prof. Elena de Momi, Politecnico di Milano (Italy)

During minimally invasive surgery, advanced image processing techniques can enhance the context awareness, e.g. during radical prostatectomy with lymphadenectomy. In this way, the surgeon will be able to detect and safely remove hidden regions, thus increasing the clinical outcome. In this talk we will present our recent advancements to develop an integrated framework for augmented reality in robotic surgery.

Can Cognitive Architectures Support Autonomy in Robotic Surgery?  Prof. David Vernon, Carnagie Mellon University, Kingali (Ruanda)

A cognitive architecture is a software framework which integrates all the elements required for a system to exhibit the abilities that are characteristic of a cognitive agent, orchestrating their operation to produce cognitive behaviour. Core cognitive abilities include perception, attention, action selection, memory, learning, reasoning, metacognition, and prospection. Ideally, a cognitive architecture also enables the agent to explain what it is doing, why it is doing it, and what it expects to happen when it does it. Here, we explore the implications of using a cognitive architecture to facilitate autonomy in robotic surgery. In addition to addressing the task of planning actions, we focus on the elements of a cognitive architecture that are needed to satisfy the requirement that a human supervisor must be able to understand the basis for actions performed autonomously by the robot, allowing the supervisor to intervene, when necessary. Fulfilling these requirements means that the cognitive architecture must enable the surgical robot to interact with its human supervisor in a timely manner. This imposes a requirement for the cognitive architecture to support advanced forms of prospection: to continually anticipate the need to act, to anticipate the outcome of these actions, to form intentions to act, and to communicate these intentions to the human supervisor before taking action. It also implies the need for the cognitive surgical robot to be able to anticipate the intentions of the human supervisor. This ability, often referred to as a theory of mind, means the robot has to take a perspective on the agent with which it is interacting, i.e., the human supervisor. The realization of this ability for collaborative interaction is one of the biggest challenges for cognitive systems research. We review progress in addressing the challenge, focussing on ways to implement theory of mind in a cognitive architecture that supports autonomy in robotic surgery.

The EU ‘AI ACT’: a “risk-based” legislation for robotic surgery, Prof. Federico Costantini (Dep. of Legal Sciences, University of Udine, IT)

The long-awaited European Union “Artificial Intelligence Act” has been recently approved (13rd of March 2024). Even though it has not been published – for this reason we still might recall it as COM(2021)206 – and despite the fact that it will come into force only after two years since its publication, it has  drawn the attention from the international community of AI experts, due to the fact that it is the first piece of legislation worldwide regulating such technologies.  This contribution aims at presenting the “AI ACT” with a focus on its most relevant features regarding robotic surgery. After a short overview on its background, which is brought by a very complex legal framework built within the last 25 years by the EU, I will offer a summary of its provisions, which are resulting from the “risk-based” approach adopted by the EU legislator. Then, I will address “high risk” AI systems, analysing both the obligations that not only manufacturers, but also providers will need to fulfil, highlighting those which are most challenging in the sector of robotic surgery. At the end I will offer a few conclusive remarks, concerns and recommendations.

The emerging European AI liability framework and the impact on medical practitioners, Prof Andrea Bertolini, Scuola Superiore San Anna (Italy)

The presentation introduces the recently modified European regulatory framework for liability arising from the use of advanced technologies, including AI. In particular it focuses on the recently approved reform to the Product Liability Directive. It then introduces briefly the proposal for an AI civil liability directive, not yet approved, and whose discussion came to a halt some months ago. Based on the assessment performed, the incentives the emerging European regulatory framework provides to the parties involved is discussed, with a focus on the liability of medical practitioners that make use of AI-based systems in (i) surgery and (ii) diagnostics

Philosophical-legal analysis of the ‘myths’ included in the AI Act about  ‘digital sovereignty’ and ‘anthropocentrism’: subsidiarity and green revolution, Prof. Andrea Favaro, Universita’ di Verona (Italy)

Ethics of autonomous systems, Prof. Sara Patuzzo, University of Verona (Italy)

While in certain medical fields, such as medical diagnostics, radiology, or anatomopathology, the AI systems utilized exhibit high levels of autonomy, in the surgical domain, the employed robots (Da Vinci® – Intuitive; Hugo™ RAS System – Medtronic; Versius – CMR Surgical) possess autonomy limited to minor adjustments from the initial input, for instance, to stabilize hand tremors or follow precise cutting lines (as with the RIO® orthopedic robot) or to align primary joysticks with surgical instruments (as in the Da Vinci® system). However, even though they are currently effectively commanded by human surgeons, who make decisions on how to act, the rapid and significant advancements in the fields of AI and robotics suggest the possibility of a future in which these robots become increasingly autonomous, potentially even replacing human surgeons. This scenario raises numerous ethical questions. Can a machine equipped with AI be considered a moral agent and thus be held accountable to us? Despite potentially being more efficient and capable than a human surgeon, how can an AI surgical robot pursue the patient’s welfare in accordance with the principle of beneficence? How should we ethically regard the fact that an AI machine does not experience emotions and empathy towards the patient within the medical context? How can ethical aspects such as autonomy and informed consent, equity, and justice be adapted? These are some of the questions that will be explored through a logical-rational method.

A critique of Floridi’s theory of machines as autonomous moral agents, Prof. Carlo Chiurco, University of Verona (Italy)

The presentation is divided in two sections. The first section aims at discussing and criticising Luciano Floridi’s theory of machines as autonomous moral agents. Given that Floridi’s argument is made of a founding, theoretical i.e. ontological part, followed by an ethical-practical one, any attempt at a critique must follow this organisation. As such, I will begin by briefly objecting about Floridi’s asserted and seemingly self-evident perfect convertibility between “being” and “data”: problems here arise on both counts – namely, concerning both such convertibility as well as its self-evidence. I will then move to a more thorough discussion of the ethical part. Floridi’s strategy allows him to bypass discussions about what is autonomy: indeed, he does not provide a direct demonstration of machines’ autonomy, but he draws such conclusion indirectly. In other words, he doesn’t say that machines are moral agents because they are autonomous – as typically found in Western philosophical tradition, which he contests – but the other way round i.e., if machines are moral gents, then they must necessarily be considered autonomous. I will first object about the legitimacy of such move, showing that categories such as desire and embodiment cannot simply be theoretically skipped, in what appears to be an extremised, lifeless – and not simply bodyless – version of Cartesian subjectivity. I will then question the notion that allows Floridi’s ethical argument to stand, namely that “preserving and improving information constitutes an ethical good”: does the legitimacy of such claim stand a closer scrutiny?

The second part will put the conclusions reached in the first part in dialogue of current attempts at regulating AI. While appreciating Floridi’s attempt at addressing the question of machines’ autotomy as pioneering, even visionary, I will show how his perfectly egalitarian perspective is highly incompatible with the far more cautious approach of UN’s “Governing AI for Humanity” and EU’s AI Act.