I am traveling to Noordwijk, Netherlands this Thursday to visit European Space Agency again and do a final presentation summarising the whole project, experimental results and suggesting future directions.
The scientific rationale of this project has been to strengthen the Evolutionary Robotics methodology which seeks the automatic design of Artificial Neural Networks as controllers for rovers (or robots in general), by making use of recent advances in global optimisation. Specifically, the research and technological roadmap included:
- Integration of the custom rover's physics simulation environment developed at the University of Plymouth with the PaGMO libraries implementing the island model for global optimisation developed by ESA-ACT
- Demonstration that such an integrated system enhances the design of neuro-robust controllers for rovers’ navigation tasks, including a complex domain such as active vision
Results
- Extension and integration of the Mars Rover Simulator developed by the university of Plymouth with PaGMO for an open source application on robotic island experiments
- Assessment of the contribution, but also the limiatations and constraints for the use of the island paradigm in robot exploration experiments
- Testing on model robustness in different environments and controller configurations
- Promising preliminary results on the evolution of active vision strategies for the integration of local (e.g., sensing sand/icy/standard ground) and distal (landmark) information in exploration tasks
The study has demonstrated the robustness of the island approach to evolve controllers for autonomous rover navigation. Comparisons between island model and single population models on different tasks show that the island model performance is statistically significantly better, by producing improved results in terms of computational performance (island model is faster) and quality of the evolved neuro-controllers. The active vision experiments also show the feasibility of the proposed methodological approach for complex tasks requiring multiple levels of decision-making and action.