In this paper, we summarize recent research enabling high-speed navigation in unknown environments for dynamic robots that perceive the world through onboard sensors. Many existing solutions to this problem guarantee safety by making the conservative assumption that any unknown portion of the map may contain an obstacle, and therefore constrain planned motions to lie entirely within known free space. In this work, we observe that safety constraints may significantly limit performance and that faster navigation is possible if the planner reasons about collision with unobserved obstacles probabilistically. Our overall approach is to use machine learning to approximate the expected costs of collision using the current state of the map and the planned trajectory. Our contribution is to demonstrate fast but safe planning using a learned function to predict future collision probabilities.
This paper presents our solution for enabling a quadrotor helicopter to autonomously navigate unstructured and unknown
indoor environments. We compare two sensor suites, specifically a laser rangefinder and a stereo camera. Laser and camera
sensors are both well-suited for recovering the helicopter's relative motion and velocity. Because they use different cues
from the environment, each sensor has its own set of advantages and limitations that are complimentary to the other sensor.
Our eventual goal is to integrate both sensors on-board a single helicopter platform, leading to the development of an
autonomous helicopter system that is robust to generic indoor environmental conditions. In this paper, we present results
in this direction, describing the key components for autonomous navigation using either of the two sensors separately.