Safe multicopter navigation in cluttered and dynamic environments

Awarded by
SLAMcore Ltd

SLAMcore Ltd is an Imperial College spin-out company specialising on Spatial AI for robots and drones. Since founded in 2016, it has received several rounds of Venture Capital, totalling to USD 10 M.


WP 1: Navigation in static environments

In this work package, the student will develop integrated model-predictive control, motion-planning and collision avoidance (online re-planning) strategies. As a basis, we will use approaches from Nils Funk’s MSc thesis, but will extend towards use of the identified nonlinear dynamics of the drone, and exploring alternative planners. For mapping using depth cameras, an adaptation of the supereight framework will be used. For accurate drone pose tracking, we will support both open-source solutions, e.g. OKVIS , developed by the PI and also available under BSD, as well as SLAMcore’s proprietary pose estimation solutions – i.e purely used as-is as an input to the algorithms developed.

WP 2: Safe motion planning in dynamic environments

This work package will extend WP1 to account for multiple and heterogeneous dynamic objects in it. We will assume known/steerable future trajectories of the other agants (e.g. other drones as part of a centrally administrated multi-robot scenarios), which we will then gradually evolve towards considering uncertainty, local information (in a limited agent neighbourhood).

WP 3: Motion tracking and prediction for sensed agents/objects

This final work package will take the navigation stack to the next level, as it will deal with agents and objects that are not under control, thus require tracking (a 3D vision problem) as well as developing data-driven models (using Machine Learning) for accurate motion prediction that includes accurate estimation of related uncertainty. The predictions can then be integrated into the control/planning/avoidance approaches from WP1 and WP2.

Delicious Twitter Digg this StumbleUpon Facebook