Event
Authors propose modular stack for onboard vision-language navigation
1 reports · 1 sources
Marco S. Tayar and other authors propose an onboard vision-language navigation stack called VLN on the Fly, which keeps instruction grounding, planning, and control as separately inspectable stages: a quantized VLM grounds the instruction to coarse-grained image cells, depth information lifts it into a 3D target, a fast B-spline planner generates feasible trajectories, and a pretrained reinforcement learning policy tracks the trajectory on multiple quadrotors and outputs motor commands. The authors report that, in a controlled indoor space, the stack completed 15 onboard flights targeting three everyday reference objects; there were also 6 trials in cluttered environments, tracking collision-free trajectories under onboard perception gating. These results come from the authors' self-reported experiments and have not yet been independently verified.
Report timeline
- VLN on the Fly: An Onboard Vision-Language Navigation Stack for Aerial Robots
The authors propose VLN on the Fly, an onboard vision-language navigation stack that keeps instruction grounding, planning, and control as separately inspectable stages: a quantized VLM grounds the instruction to a coarse image cell, depth lifts it to a 3D goal, a fast B-spline planner generates a feasible trajectory, and a pretrained reinforcement learning policy tracks the trajectory and outputs motor commands across multiple quadrotors.
arXiv cs.RO ·