Lamarckian Inheritance Improves Robot Evolution in Dynamic Environments

TL;DR

It is demonstrated that Lamarckian systems outperform Darwinian ones in adaptability and efficiency, particularly in challenging conditions, and could significantly advance autonomous system design, highlighting the potential for more adaptable and robust robotic solutions in complex, real-world applications.

Abstract

This study explores the integration of Lamarckian system into evolutionary robotics (ER), comparing it with the traditional Darwinian model across various environments. By adopting Lamarckian principles, where robots inherit learned traits, alongside Darwinian learning without inheritance, we investigate adaptation in dynamic settings. Our research, conducted in six distinct environmental setups, demonstrates that Lamarckian systems outperform Darwinian ones in adaptability and efficiency, particularly in challenging conditions. Our analysis highlights the critical role of the interplay between controller \& morphological evolution and environment adaptation, with parent-offspring similarities and newborn \&survivors before and after learning providing insights into the effectiveness of trait inheritance. Our findings suggest Lamarckian principles could significantly advance autonomous system design, highlighting the potential for more adaptable and robust robotic solutions in complex, real-world applications. These theoretical insights were validated using real physical robots, bridging the gap between simulation and practical application.

Figures (15)

• Figure 1: Overview of environmental conditions and experimental Setups. Static setup involves no environmental changes throughout the entire evolution process and is tested on two terrains (Flat and Rugged). Dynamic slow and Dynamic fast configurations involve environmental conditions changing with frequencies of 2 and 5, respectively. In both cases, two methods are employed: one starting with flat terrain (Flat_2 or Flat_5) and the other with rugged terrain (Rugged_2 or Rugged_5). All setups undergo two evolution frameworks, the Darwinian system and the Lamarckian system, resulting in a total of 6 environmental setups, 12 experiments.
• Figure 2: (a) Mean (lines) and maximum (dots) fitness over 30 generations in 6 environmental setups. The red dotted rectangles show the transition from easy flat terrain to complex rugged terrain. The black dotted rectangles are from complex to easy terrain. (b) Progression of the learning delta throughout evolution averaged over 10 runs. The bands indicate the 95% confidence intervals (Sample Mean $\pm$ t-value $\times$ Standard Error).
• Figure 3: Transferability during environmental transitions: (A) Mean of the ratios from easy to complex terrain. (B) Mean of the ratios from complex to easy terrain. (C) Mean of the ratios for every transition.
• Figure 4: Mean morphological similarity (a) and controller similarity (b) over generations in 6 environmental setups.
• Figure 5: Correlation between fitness & controller similarity (a) and morphological dissimilarity (b) in 6 environmental setups. The dots are the aggregated individuals across all runs. Orange dots represent the Lamarckian system, while blue dots represent the Darwinian system.