This Wednesday at 3pm at CB11.09.215, Jen will be doing a practice run of her CA3.

Title: Sparse belief space planning for active perception

Abstract:  To make smart and safe decisions in dynamic and unstructured environments, robots must have accurate estimates of target systems that are relevant to their task. These target systems may be other entities, environmental features or landmarks. Active perception, or the planning of sensing trajectories to reduce uncertainty in estimates of target states is then crucial for good decision making. Estimates are often represented as probability density functions over the space of physical target states, or beliefs. Planning over the space of possible such probability distributions is known as belief space planning and is a popular approach for active perception, as it facilitates the direct addressal of uncertainty while planning. However, belief space is a continuous space of prohibitively large dimension. It is often infeasible to find optimal sensing trajectories in the real-time requirements of robotics by planning in belief space. As such, research in recent decades has focused on finding suboptimal solutions to belief space planning problems in feasible time frames. Motivated by this research, this thesis interrogates how structures inherent to common active perception problems – active target tracking and environmental monitoring – can be exploited for efficient belief space planning. For each problem setting we show how structure admits sparse belief spaces for planning sensing trajectories. Evaluations of proposed methods demonstrate that these sparse belief spaces lower the computational complexity of planning while maintaining rich information regarding the target state. Moreover, performance guarantees associated with planning over the proposed belief spaces are provided.  

Afterwards, if time permits, lets read through Volume 4 of Field Robotics (link below):
https://mail.journalfieldrobotics.org/Field_Robotics/Volume_4.html

See you all there!