Flight Through Cluttered Environments
| Through a collaborative ONR-MURI project with MIT, CMU, Stanford, and NYU, Huai-Ti Lin (former postdoc now at Janelia Farms HHMI), Ivo Ros and I carried out studies of pigeons flying through a virtual forest of vertical PVC poles. This project seeks to understand how pigeons plan and navigate flight paths to avoid obstacles similar to those encountered in their natural environment, helping to achieve more robust real-time control of unmanned aerial vehicles (UAVs) using insights gained from studies of pigeon visuomotor control of flight and maneuvering. 3D kinematics track motion of the pigeon's head and body using active LED markers, with respect to known pole locations that are randomized across trials. (Lin et al. J. Roy. Soc. Interface, 2014) |
Ivo Ros and Huai-Ti Lin |
| Pigeons are trained to fly through the cluttered flight course located between two perches, allowing evaluation of visual cues (looming, time-to-collision, gap size and gap location) that the pigeons may use to guide their flight. | |
|
From these data, we developed a navigational model using proportional and derivative control of obstacle locations and available gaps. Simulations based on a simple control scheme to select the widest gap between nearby obstacles accurately predicted ~70% of the observed flight paths taken by the pigeons through the virtual forest, with deviations largely dependent on L:R bifurcations in simulated vs observed paths past an obstacle. Onboard video camera recordings provide a sense of what it's like to fly with the pigeon through such an environment (NB: pigeon's visual field >300deg substantially exceeds the camera's 60deg field. And, it is likely that pigeons rely on high resolution lateral fovea and optical flow in the lateral fields of their eyes to guide their flight). Ongoing work seeks to explore how well pigeons select and learn a particular flight path and what other visual cues may improve model simulations. We are also examining whether our 'widest gap selection model' can accurately predict the paths pigeons take when maneuvering past horizontal obstacles (vs vertical obstacles).
|
|
|
|
|
