Modeling Complex Biological and Mechanical Movements: Applications to Animal Locomotion and Gesture Classification in Robotic Surgery
Johns Hopkins University
Mutual interaction between biology and robots can significantly benefit both fields. The richness and diversity in animal locomotion and movement provides an extensive resource for inspiration in engineering design of robots. On the other hand, bio-mimetic and bio-inspired robots play a critical role in testing hypotheses in biology and neuromechanics. Modeling complex biological and mechanical movements is at the core of this mutual interaction. Models and analytical tools are required for decoding and analysis of behavior in biological and mechanical systems, both at low level (sensory systems and control) and high level (activity recognition). This dissertation is focused on modeling approaches for biological and mechanical movements. We first primarily focus on physics-based template modeling to answer a long-standing question in animal locomotion: why do animals often produce substantial forces in directions that do not directly contribute to movement? We examine the weakly electric knifefish, a well-suited model system to investigate the relationship between mutually opposing forces and locomotor control. We use slow-motion videography to study the ribbon-fin motion and develop a physics-based template model at the task-level for tracking behavior. Using the developed physics-based model integrated with experiments with a biomimetic robot, we demonstrate that the production and differential control of mutually opposing forces is a strategy that generates passive stabilization while simultaneously enhancing maneuverability, thereby simplifies neural control. The second part of this work aims to propose a more general data-driven system-theoretic framework for decoding complex behaviors. Specifically we introduce a new class of linear time-invariant dynamical systems with sparse inputs (LDS-SI). In the proposed framework, at each time instant, the input to the system is sparse with respect to a dictionary of inputs. In the context of complex behaviors, the dictionary may represent the dictionary of inputs for all possible simple behaviors. We propose a convex optimization formulation for the state estimation with unknown inputs in LDS-SI. We derive sufficient conditions for the perfect joint recovery and explore the results with simulation. We demonstrate the power of the proposed framework in the analysis of complex gestures in robotic surgery. Results are better than state-of-the-art methods in joint segmentation and classification of surgical gestures in a dataset of suturing task trials performed by different surgeons.
Bio-inspired Robotics, Linear Dynamical Systems with Sparse Inputs