A new arXiv announcement, VIGOR-MPC, takes on the problem of controlling an agent when it perceives the environment only through onboard vision. The abstract notes that such perception is challenging because it produces noisy, variable-length, and partially occluded context, while onboard compute must be shared between perception and control.

The proposed method, Vision-Guided Objective Regulation MPC, appears to extend model predictive control by regulating objectives derived from visual input. However, the source is an announcement with a truncated abstract, so no details on the algorithm's implementation or experimental results are available in the provided text.

Given the limited source, we can only report the problem framing and the name of the approach. The paper clearly targets the intersection of vision and control under real-world constraints, but specifics remain unavailable from the abstract excerpt.