We investigate collaborative coverage with a self-organizing swarm of minimalist robots. The inspection of the compressor section of a (jet) turbine engine is a motivating case study (see IROS’06 and BSSAC’07 for a high-level overview).
Based on a simple, randomized solution in which robots do not collaborate (ISER’04), we introduce local communication among the robots in order to improve dispersion in the environment and thus speed up coverage progress (ISER’06).
As the individual robots are extremely unreliable due to sensor and actuator noise, we capture the dynamics of the swarm using probabilistic microscopic and macroscopic models (DARS’04, ICRA’05, DARS’06), which can serve as a prediction and design tool (SYROCO’06).
We contrast this approach with deliberative algorithms that theoretically provide complete coverage (ISER’06), but decay to probabilistic completeness due to sensor and actuator noise (ICRA’07). Currently, we are investigating collaborative approaches, where information about task progress is shared using a custom developed radio module running TinyOS, and offline algorithms that provided near-optimal coverage paths to the robot team.
Real robot experiments are carried out with a swarm of up to 40 Alice miniature robots, developed at the Autonomous Systems Laboratory by Gilles Caprari. For actual inspection and enhanced navigation/localization, we equipped the Alice platform with a miniature camera that allows us capturing 30×30 pixels RGB image at around 2Hz (video). We also use the realistic simulator Webots to systematically study the impact of different hardware constraints.
Team and Collaborators
Sponsors and Research Period
NASA Glenn research Center (till September 2003)
Caltech Center for Neuromorphic Systems Engineering under the US NSF Cooperative Agreement ERC-9402726 (till September 2003).
Swiss National Science Foundation contract Nr. PP002-68647 (from October 2003-October 2007)
- Video broadcasted over the radio-link (AVI)
- Collective Inspection with a swarm of 20 Alice miniature robots (Flash) (AVI)
- Collective Inspection in embodied simulation (Webots)
This document contains supplementary material to the paper “Distributed Coverage: From Deterministic to Probabilistic Models”, submitted to the International Conference on Robotics and Automation, Rome, 2006.
- Inspection Progress for 1 to 10 robots with wheel-slip 0.1 (100 experiments each)
- Inspection Progress for 1 to 10 robots with wheel-slip 0.5 (100 experiments each)
- Time to failure for wheel slip of 0.1 and 0.5 (6000 experiments each)
- MATLAB source code for multi-robot spanning tree coverage and visualization
- Supplementary video (Macromedia Flash)