Head of Learning
Engineering expertise is key to creating AI that works in the physical world. And where does this expertise come from? At the most fundamental level, it means understanding a system's goals and constraints while representing its complex dynamics and physical realities through figures and variables.
Composabl's no-code platform empowers you, the agent builder, to translate expertise into guidance that trains specialized AI "skill agents" for complex processes, all through a user-friendly interface.
Using the platform, agent builders input goals, constraints, and success based on real-world expertise; under the hood, Composabl translates these inputs into reward functions. For each skill agent, a deep reinforcement learning algorithm practices in simulation and, through repetition and experimentation, develops expertise by learning how to achieve the goals while staying within the constraints until the success criteria is reached.
Goals define what a skill agent should do.
Agent builders select a variable from within the simulation or real system, then have the option to choose between three goal objectives for that variable: maintain (which directs the system to learn to track to a set point), maximize, or minimize.
When a goal is set, the system will then focus in training sessions on learning how to maximize, minimize or maintain the variable as directed.
Constraints set the rules of the game, the boundaries for the skill agent. Agent builders can utilize two criteria to set constraints: avoid and terminate.
Avoid criteria. The skill agent learns to stay away from a given behavior (represented as a numerical range) through withholding the reward.
Terminate criteria. When the actions of the skill agent lead to certain conditions within a variable, the skill agent has failed and must stop and start a new episode.
Success criteria are the opposite of constraints – they tellthe skill agent when it’s doing something right. As with constraints, Composablusers use two different methods to direct agent training with success criteria.
Approach criteria. Approach is the compliment to avoid. The skill agent learns to get close to a specified value by getting increased reward.
Succeed criteria. Succeed is the compliment to terminate. When the success criteria are achieved, the session ends, and a new one begins so that the skill agent can keep practicing and learn to win every time.
Goals, constraints, and success criteria transform complex AI agent building into an intuitive process that eliminates the need for programming expertise.
By simply defining what you want the agent to achieve and what conditions to respect, you directly apply your domain knowledge without coding. This approach reduces development time while ensuring AI agents embody real-world operational wisdom, bridging the gap between technical AI capabilities and practical industrial knowledge in complex physical environments.
Want to see how this works in context of the entire platform? Register for a LIVE platform walkthrough here.
