Yanran Xu, Klaus-Peter Zauner, Danesh Tarapore
Agents Interaction & Complexity, University of Southampton, Southampton, U.K.
Accepted to IROS 2025 🎉
Reliable perception is not always available so robots may keep colliding with the same obstacles;
Accurate localization is not always available so robots may collide with the same obstacle even if they try to avoid it by estimating the robot-obstacle relative position;
Robots may get trapped in perception-degraded clustered forest scenarios when reliable perception and accurate localization are not available.
A probing-driven navigation framework to prevent robots from remaining trapped when obstacles cannot be reliably perceived or even in perception-denied scenarios;
The design of an abstract forest simulation environment to develop and systematically evaluate our probing-driven navigation framework;
Experimental validation of the proposed framework both in simulation and in real-world forest scenarios with two different robot platforms.
The blue point and arrow mark the starting position and orientation;
The yellow line indicates the destination line;
The inset image highlights the main obstacles with red lines.
Representative examples from 500 abstract forest simulation environments;
Each environment spans a 50 x 50 m area with a tree density of 80 trees/hectare, where each tree has a radius of 0.25m;
Black lines indicate fallen branches, green circles represent stating trees and red dots mark the locations where the robot becomes immobilized or slips on rough terrain accidentally ;
The green line shows the robot’s trajectory. The red star indicates the starting point and the black arrow denotes the designated heading.
1000 experiments for each algorithm;
500 random generated simulation environments;
2 experiments in each environment with different random seed.
80 real-world experiments;
2 distinct trial scenarios;
2 rover platforms with different sizes;
45 m shortest distance from the starting point to the destination line (paved pathway);
Multiple challenging scenarios during the trial;
1000 experiments in 500 randomly generated simulation environments;
Limited effects on performance when the size of robots changes;
MiniRover: 200 mm x 200 mm
ForestRover: 400 mm x 400 mm
1000 experiments in 500 randomly generated simulation environments;
Performance evaluated with Forest-Pledge;
Performance is better within the range between 10 to 50 degrees.
Performance is robust within the range between 10 to 50 degrees.
Gray bars: Escape time
Brown bars: Entrapment trials
Red Vertical lines: Human interventions
Dense Obstruction
Extended Maze
Dense Obstruction
Extended Maze
Forest-Pledge
Lévy walk
Forest-Pledge
Lévy walk
Forest-Pledge
Lévy walk
Forest-Pledge
Lévy walk