Laboratory for Computational Sensing + Robotics, Johns Hopkins University

Medical imaging is increasingly moving beyond diagnosis to become an integral component of therapy. Terry Peters illustrates this transition with two very different applications from his group: image-guided mitral valve repair and Intratumoural Modulated electric-field Therapy (IMT) for brain tumours. For mitral valve intervention, multimodal imaging, 3D visualisation and patient-specific dynamic models improve navigation and allow alternative repair strategies to be rehearsed and evaluated before intervention. For glioma, patient imaging is used to plan electrode trajectories and optimise three-dimensional electric-field distributions to maximise tumour coverage while sparing surrounding structures, followed by image-guided implantation and verification; pre-clinical validation of IMT uses bioluminescence imaging and MRI to demonstrate its efficacy. Together the examples show imaging as part of planning, optimising, guiding and evaluating patient-specific therapy.

Aerial robots have become ubiquitous but, like most robots, still struggle to operate at high speed in unstructured, cramped environments. Mark Mueller presents two ideas for creating more capable aerial systems. The first designs the control and mechanical properties of a system simultaneously, applied to carrying a load with multiple aerial robots: rather than fixing open-loop characteristics such as thrust-to-weight ratio first and designing the feedback controller afterwards, it uses insights from robust control to design both together, based on models of the system dynamics and external disturbances. The second is a learning approach for robust and adaptable quadcopter control strategies, demonstrated in experiments to search over very large parametric uncertainty (inertia spanning two orders of magnitude) and to respond almost instantaneously to parameter changes such as offsets in centre of mass or reduced motor effectiveness.

The expanding reach of robotics and AI brings acute reliability challenges under limited information, rapidly changing conditions and close human interaction. Safety filters, runtime schemes that monitor a system's candidate actions and intervene to preempt catastrophic failures, have attracted wide interest, but the restrictive assumptions of model-based approaches and the weaker guarantees of data-driven alternatives have reinforced the perception of a fundamental safety-performance tradeoff. Jaime Fernández Fisac presents theoretical and experimental evidence against that view. Building on a unified understanding of safety filter mechanics, he introduces a general class of neural safety filters that extend the rigorous guarantees of safety-critical control beyond the reach of classical synthesis. Safety over a specified operational design domain (ODD) is cast as an adversarial reinforcement learning problem, co-training the system's controller against the worst environment realisations consistent with the ODD. These neural approximations enable certified safety filters in domains from legged locomotion to high-speed racing, including the synthesis of control barrier functions for general nonlinear systems with only black-box access to their dynamics. Wrapped around safety-agnostic planning and learning pipelines, the filters achieve near-optimal task performance in safety-critical ODDs while provably enforcing zero catastrophic failures. The talk then turns to human-centred systems, where safety requires steering interactions away from downstream harms, from anticipating black-swan road encounters to safeguarding novice human operators, and closes with early results on dynamic guardrails for generative AI assistants and open challenges for safety-critical intelligence beyond robotics.