Lamarckian Inheritance in Dynamic Environments: Key Variables Affecting Robot Evolution
A new study on arXiv (2605.15769) investigates Lamarckian inheritance in evolutionary robotics, where learned controller parameters are passed from parent to offspring. The research addresses conflicting evidence in the field: traditional evolutionary theory often finds Lamarckian inheritance unhelpful, but recent robotics studies show performance gains. The authors hypothesize that prior work omitted key variables related to dynamic environments. They demonstrate that the benefit of Lamarckian inheritance depends on two factors: how conflicting environmental changes are to robot control, and an unspecified second variable. The study combines morphology optimization (evolution) with controller optimization (lifetime learning) to co-optimize robot body and brain.
Key facts
- Study published on arXiv with ID 2605.15769
- Investigates Lamarckian inheritance in evolutionary robotics
- Combines morphology optimization as evolution with controller optimization as lifetime learning
- Finds benefit depends on two variables related to dynamic environments
- First variable: how conflicting environmental changes are to robot control
- Addresses conflicting evidence in existing literature
- Traditional evolutionary theory suggests Lamarckian inheritance lacks benefit
- Recent studies in evolutionary robotics indicate it can improve performance
Entities
Institutions
- arXiv