- Threshold dynamics of SAIRS epidemic mannequin with Semi-Markov switching
Authors: Stefania Ottaviano
Summary: We examine the edge dynamics of a stochastic SAIRS-type mannequin with vaccination, the place the function of asymptomatic and symptomatic infectious people is explicitly thought of within the epidemic dynamics. Within the mannequin, the values of the illness transmission charge might swap between completely different ranges beneath the impact of a semi-Markov course of. We offer ample circumstances guaranteeing the virtually certainly epidemic extinction and persistence in time imply. Within the case of illness persistence, we examine the omega-limit set of the system and provides ample circumstances for the existence and uniqueness of an invariant chance measure.
2. Studying Reactive and Predictive Differentiable Controllers for Switching Linear Dynamical Fashions
Authors: Saumya Saxena, Alex LaGrassa, Oliver Kroemer
Summary: People leverage the dynamics of the setting and their very own our bodies to perform difficult duties resembling greedy an object whereas strolling previous it or pushing off a wall to show a nook. Such duties typically contain switching dynamics because the robotic makes and breaks contact. Studying these dynamics is a difficult drawback and vulnerable to mannequin inaccuracies, particularly close to contact areas. On this work, we current a framework for studying composite dynamical behaviors from professional demonstrations. We study a switching linear dynamical mannequin with contacts encoded in switching circumstances as an in depth approximation of our system dynamics. We then use discrete-time LQR because the differentiable coverage class for data-efficient studying of management to develop a management technique that operates over a number of dynamical modes and takes into consideration discontinuities because of contact. Along with predicting interactions with the setting, our coverage successfully reacts to inaccurate predictions resembling unanticipated contacts. Via simulation and actual world experiments, we exhibit generalization of realized behaviors to completely different situations and robustness to mannequin inaccuracies throughout execution.
3. Dynamically Switching Human Prediction Fashions for Environment friendly Planning
Authors: Arjun Sripathy, Andreea Bobu, Daniel S. Brown, Anca D. Dragan
Summary: As environments involving each robots and people grow to be more and more widespread, so does the necessity to account for folks throughout planning. To plan successfully, robots should be capable of reply to and generally affect what people do. This requires a human mannequin which predicts future human actions. A easy mannequin might assume the human will proceed what they did beforehand; a extra advanced one would possibly predict that the human will act optimally, disregarding the robotic; whereas an much more advanced one would possibly seize the robotic’s potential to affect the human. These fashions make completely different trade-offs between computational time and efficiency of the ensuing robotic plan. Utilizing just one mannequin of the human both wastes computational sources or is unable to deal with vital conditions. On this work, we give the robotic entry to a collection of human fashions and allow it to evaluate the performance-computation trade-off on-line. By estimating how an alternate mannequin might enhance human prediction and the way which will translate to efficiency acquire, the robotic can dynamically swap human fashions each time the extra computation is justified. Our experiments in a driving simulator showcase how the robotic can obtain efficiency similar to at all times utilizing the perfect human mannequin, however with tremendously diminished computation.