Truclo
Introducing a framework to create AI agents that can understand human instructions and perform actions in open-ended settings.
About Truclo
Introduction
Artificial intelligence has made significant strides in recent years, but one of the biggest challenges remains developing agents that can operate in open-ended environments with minimal human guidance. Truclo is tackling this challenge by creating AI agents capable of understanding instructions, learning from experience, and dynamically adapting to new scenarios within video game worlds.
Unlike traditional AI systems that rely on predefined behaviors or rigid rule sets, these agents are designed to generalize knowledge and act independently based on learned objectives. By combining imitation learning, reinforcement learning, and human feedback, Truclo has developed an innovative approach to training AI in complex, interactive environments.
Framework
Truclo has developed an advanced AI framework designed to create agents that can interact with open-ended video game worlds. This framework integrates three key components: imitation learning, reinforcement learning, and human feedback. By combining these techniques, Truclo enables AI to understand instructions, learn from experience, and adapt to new tasks dynamically.
1 - Imitation Learning: Establishing a Foundation
2 - Reinforcement Learning: Optimizing Actions
3 - Generalization: Applying Knowledge to New Tasks
4 - Human Feedback: Refining Behavior
Find out more below:
The Old Playhouse
We created a virtual "playhouse" with hundreds of recognisable objects and randomised configurations. Designed for simple and safe research, the interface includes a chat for unconstrained communication.
Our framework begins with people interacting with other people in the video game world. Using imitation learning, we imbued agents with a broad but unrefined set of behaviours. This "behaviour prior" is crucial for enabling interactions that can be judged by humans. Without this initial imitation phase, agents are entirely random and virtually impossible to interact with. Further human judgement of the agent’s behaviour and optimisation of these judgements by reinforcement learning (RL) produces better agents, which can then be improved again.
Find out more about the old playhouse here:
The New Playhouse
This is what one of the divisions of the new playhouse looks like, the main room. The old playhouse will be replaced by this space and all the other divisions that are in construction.
Why?
We decided it was time to add more agents to the space, more color and more complex objects so that agents can perform more complex tasks.
There are also computers in this new playhouse, where agents will actually be able to truly interact with them and access the web.
Each agent will also have its own wallet, and will be free to do whatever they want without any limitations. This will only be done after they completely understand real human behaviour.
Explore the new playhouse here:
