VirtualEnaction - A Platform for Systemic Neuroscience Simulation

Nicolas Denoyelle, Florian Pouget, Thierry Viéville, Frédéric Alexandre2014 · DOI

TL;DR

This paper introduces VirtualEnaction, an open-source platform that leverages the Minecraft game engine to simulate complex, biologically plausible brain models interacting with a dynamic environment. It addresses the need for realistic, controllable, and affordable tools for systemic neuroscience research, enabling the testing of computational models in survival-oriented scenarios.

What Problem Does This Paper Address?

Studying the brain as a whole, rather than isolated components, presents a significant challenge. Researchers need tools that can realistically simulate complex brain models and allow for experimentation within dynamic environments. Traditional robotic setups are often too expensive and difficult to manage for long-term, complex experiments. Furthermore, existing simulation tools may lack the necessary realism or control for validating functional brain models, especially when considering emergent behaviors arising from system-wide interactions. The authors highlight the need for a platform that bridges the gap between computational neuroscience and realistic experimental paradigms, allowing for both in-depth analysis of brain models and the design of artificial autonomous systems.

Methods

VirtualEnaction is built upon the Minecraft game engine, chosen for its accessible 3D environment, block-based world, and existing user base. The platform provides a "brainy-bot" that embodies a computational model of the brain. This bot interacts with the Minecraft environment through a simplified sensory-motor loop, receiving inputs like visual, tactile, and olfactory cues, and outputting motor commands for movement, gestures, and actions.

Key methodological choices include:

  • "Look-and-move" paradigm: The simulation runs at a slower-than-real-time pace, allowing for detailed observation and analysis without the constraints of real-time robotic control.
  • Symbolic motor commands: Instead of low-level motor control, the platform focuses on higher-level commands (e.g., "stand-up," "turn 90° rightward"), enabling the study of action selection and motor planning.
  • Complex but not necessarily natural visual environment: While the visual environment is rich with textures and moving objects, it doesn't aim for photorealistic natural scenes, focusing instead on cues relevant to survival tasks.
  • Open-source and modular design: The platform is freely available and designed with a plugin architecture, allowing for easy integration of new functionalities and extensions.

Datasets

Not discussed in the paper. The platform is designed to create its own experimental environments within Minecraft.

Key Insights and Ideas

  • Minecraft as a neuroscience sandbox: The paper demonstrates that a popular video game can be repurposed as a powerful and accessible tool for complex scientific simulation, democratizing access to sophisticated experimental setups.
  • Survival as a unifying paradigm: Framing experiments around survival behaviors provides a natural and biologically relevant context for testing brain models, encompassing a wide range of cognitive functions.
  • Bridging the gap between computational and experimental neuroscience: VirtualEnaction offers a way to perform "digital experiments" that closely mimic real-world biological experiments, allowing for more rigorous validation of computational models.
  • The importance of "digital experiments": The authors advocate for a shift in methodology where computational models are tested extensively in silico before confronting them with biological reality, mirroring practices in engineering.
  • Modular design for extensibility: The plugin-based architecture is crucial for allowing the platform to evolve and accommodate diverse research needs and future extensions.

Key Results

The paper doesn't present specific quantitative results from experiments conducted on the platform. Instead, it focuses on detailing the platform's architecture, capabilities, and its potential applications. The key "result" is the successful development and release of the VirtualEnaction platform itself, which is described as being in use for preliminary digital experiments in areas like Pavlovian conditioning and decision-making.

Limitations and Future Work

Limitations:

  • Slower-than-real-time simulation: While enabling detailed analysis, this limits direct real-time interaction for human users.
  • Simplified embodiment: The description of the bot's anatomy and motor control is functional rather than highly detailed.
  • Focus on survival: While a strength, this might limit its applicability to research questions not directly related to survival.

Future Work:

  • Adding noise and degradation: The platform is designed to easily incorporate noise into sensory inputs or motor outputs to simulate real-world imperfections.
  • Developing new cues and interactions: The modular design allows for the creation of new environmental elements and interaction types.
  • Implementing middleware for cognitive functionalities: The authors mention ongoing development of modules for basic cognitive functions beyond survival.
  • Exploring bio-inspired architectures for robotics: The platform could be used to compare the effectiveness of brain-inspired control strategies against traditional robotics approaches.
  • Extending the platform: The modular architecture is expected to lead to numerous extensions in the near future.

Summary updated

Abstract

Considering the experimental study of systemic models of the brain as a whole (in contrast to models of one brain area or aspect), there is a real need for tools designed to realistically simulate these models and to experiment them. We explain here why a robotic setup is not necessarily the best choice, and what are the general requirements for such a bench-marking platform. A step further, we describe an effective solution, freely available on line and already in use to validate functional models of the brain. This solution is a digital platform where the brainy-bot implementing the model to study is embedded in a simplified but realistic controlled environment. From visual, tactile and olfactory input, to body, arm and eye motor command, in addition to vital somesthetic cues, complex survival behaviors can be experimented. The platform is also complemented with algorithmic high-level cognitive modules, making the job of building biologically plausible bots easier.
BibTeX
@misc{denoyelle2014virtualenaction,
  author = {Nicolas Denoyelle and Florian Pouget and Thierry Viéville and Frédéric Alexandre},
  title = {{VirtualEnaction - A Platform for Systemic Neuroscience Simulation}},
  year = {2014},
  doi = {10.5220/0005166701550163},
}
Figure