Skip to content
PaperFren

Robot learning

Should newborn robots inherit what their parents learned?

Jelisavcic M, Glette K, Haasdijk E, et al. · Frontiers in robotics and AI · 2019

Open access · cc by · source: Europe PMC

Letting simulated robots inherit the gait controllers their parents had learned made offspring walk faster than starting from inherited-but-unlearned controllers, especially when learning time was short.

Save to libraryView originalAsk PaperFrenStudy view

Study at a glance

Design
Computational / modelling — Simulated evolution of modular robot bodies and CPG/CPPN controllers in Revolve/Gazebo; Darwinian vs Lamarckian controller inheritance compared over five lineages and three HyperNEAT learning budgets.
N
Five lineages per condition, each with 20 robots per generation for 10 generations (200 robots per lineage); no single analytic N.
Population
Simulated modular robots built from RoboGen-style components
Outcome
Locomotion speed (fitness) after lifetime learning; parent-offspring morphological similarity; retention of parental controller material

Structured fields used in claim comparison tables when every cited study has a complete layer.

Key findings

Lamarckian inheritance produced faster robots across generations, with the biggest advantage under small and medium learning budgets; with the largest budget the Darwinian system mostly caught up. The benefit of inheriting a learned controller was larger when a child's body was more similar to its parents'. Lamarckian runs also converged on more consistent body shapes, and offspring kept more of their parents' controller structure. A few body types still learned better from scratch.

Methodology

In simulation, robot bodies built from bricks and motorised hinges were evolved for walking speed, and every new robot had a short learning period (using the HyperNEAT algorithm) to tune its controller. The controller combined a body-specific oscillator network with a body-independent network that could be passed between different bodies. In the Darwinian version, offspring received the controllers their parents were born with; in the Lamarckian version, they received the controllers their parents had learned. Both versions were run on the same five starting populations with small, medium and large learning budgets.

Limitations

Everything happens in one simulator with one task (straight-line speed), open-loop controllers without sensory feedback, and small populations of 20 over 10 generations, so results may not transfer to real hardware or other tasks; the authors say results are based on one system only. Comparisons rest mainly on plotted trends across only five lineages rather than formal significance tests of the main performance difference. The text supplied here lacks the paper's introduction and related-work sections.

How this study connects

Role on claims

Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.

  • SupportsRobot learning from rewards, demonstrations and humansconcept

    Curiosity, inheritance and few-shot examples are viable but lightly tested.

    Alternative learning routes also work in constrained settings: inheriting learned controllers (Lamarckian evolution) produced faster simulated robots under small learning budgets, a curiosity-driven humanoid explored more evenly than random agents, and a neuromorphic robot learned to turn from ten positive examples.

    Evidence for the claim as stated.

Related papers in this topic

Same topic cluster — not a recommendation engine.