Far-Field Image-Based Traversability Mapping for a Priori Unknown Natural Environments
Ethan Fahnestock, Erick Fuentes, Samuel Prentice, Vasileios Vasilopoulos, Philip Osteen, Thomas Howard, Nicholas Roy
AI summary
Problem
Robots navigating unmapped environments rely on short-range sensors, forcing planners to assign fixed costs to distant unknown space. This limits long-range path planning and often leads to inefficient routes when traversable terrain lies just beyond sensor range.
Approach
FITAM learns to correlate distant RGB image slices with terrain traversal costs in a self-supervised manner, projecting these predictions into a polar grid and fusing them over time to extend the robot's costmap horizon.
Key results
- Self-supervised learning of far-field cost predictions without range sensors or flat ground assumptions
- Improved navigation efficiency in large-scale simulated environments
- Successful real-world deployment on a Warthog robot in forest terrain
- Reliable cost estimation extended to distances up to 100 meters
Why it matters
Enables efficient long-range navigation for field robots in unstructured, GPS-denied, or rapidly changing environments where traditional proximate sensing fails.
Abstract
While navigating unknown environments, robots rely primarily on proximate features for guidance in decision making, such as depth information from lidar to build a costmap, or local semantic information from images. The limited range over which these features can be used may result in poor robot behavior when assumptions about the cost of the map beyond the range of proximate features misguide the robot. Integrating “far-field” image features that originate beyond these proximate features into the mapping pipeline has the promise of enabling more intelligent navigation through unknown terrain. To navigate with far-field features, key challenges must be overcome. As far- field features are typically too distant to localize precisely, they are difficult to place in a map. Additionally, the large distance between the robot and these features makes connecting these features to their navigation implications difficult. We propose FITAM, an approach that learns to use far-field features to predict costs to guide navigation through unknown environments in a self-supervised manner. Unlike previous work, our approach does not rely on flat ground plane assumptions or range sensors to localize observations. We demonstrate the benefits of our approach through simulated trials and real-world deployment on a Clearpath Robotics Warthog navigating through a forest environment. Code is available at github.com/efahnestock/fitam.