MAVEN and EDGAR are instances of avatar robotic systems developed at the BeingThere Centre. The primary focus of the initiative is to enhance Human-Avatar experience, whilst reducing the cognitive load of the remote human inhabitant. The approach was to imbue systems with control and decision support algorithms to allow the human to focus on the task of interactions . The system adopts modular design architecture in both the hardware and software design. This allows for easy customization of the system to suit the needs of various applications, built upon the basic needs of telepresence and robotic functions. Intrinsic to the design are features that address the need for safety and continuous interactions in the presence of short duration communication failures. In terms of physical development are MAVEN and EDGAR. The former is a mobile platform with a flat display for gesture whilst EDGAR provides an enhanced physical 3D humanoid form for gesture display.
The feature extraction module extracts a 2Kb compact descriptor from an image to describe its visual characteristic. A fast detector BFLOG is adopted in feature detection part. The total extraction procedure costs about 200ms in modern CPU.
The index module aims to build index of database for fast and efficient retrieval. ROSE_CDVS lib provides 4 the database capacity: 1K, 10K, 100K, 1M. The 1K package is free, but others need to charge. Multi-threading and incremental indexing are supported.
Image retrieval module finds the similarity images of query from the index of database. At database size 1K, 10K and 100K, a single retrieval can be finished in 300ms. At database size 1M, by adopting a fast MBIT search algorithm, a single retrieval can be finished in 500ms.