We present numerically derived orbits and mass estimates for the inner Saturnian satellites, Atlas, Prometheus, Pandora, Janus, and Epimetheus from a fit to 2580 new Cassini Imaging Science Subsystem astrometric observations spanning 2004 February to 2013 August. The observations are provided as machine-readable and Virtual Observatory tables. We estimate GM_Atlas_=(0.384+/-0.001)x10^-3^km^3^/s^2^, a value 13% smaller than the previously published estimate but with an order of magnitude reduction in the uncertainty. We also find GM_Prometheus_=(10.677+/-0.006)x10^-3^km^3^/s^2^, GM_Pandora_=(9.133+/-0.009)x10^-3^km^3^/s^2^, GM_Janus_=(126.51+/-0.03)x10^-3^km^3^/s^2^, and GM_Epimetheus_=(35.110+/-0.009)x10^-3^km^3^/s^2^, consistent with previously published values, but also with significant reductions in uncertainties. We show that Atlas is currently librating in both the 54:53 co-rotation-eccentricity resonance (CER) and the 54:53 inner Lindblad (ILR) resonance with Prometheus, making it the latest example of a coupled CER-ILR system, in common with the Saturnian satellites Anthe, Aegaeon, and Methone, and possibly Neptune's ring arcs. We further demonstrate that Atlas's orbit is chaotic, with a Lyapunov time of ~10years, and show that its chaotic behavior is a direct consequence of the coupled resonant interaction with Prometheus, rather than being an indirect effect of the known chaotic interaction between Prometheus and Pandora. We provide an updated analysis of the second-order resonant perturbations involving Prometheus, Pandora, and Epimetheus based on the new observations, showing that these resonant arguments are librating only when Epimetheus is the innermost of the co-orbital pair, Janus and Epimetheus. We also find evidence that the known chaotic changes in the orbits of Prometheus and Pandora are not confined to times of apse anti-alignment.