Which machine learning approach allows robots to learn from the consequences of their actions?

Reinforcement learning is the machine learning approach that enables robots and other agents to learn from the consequences of their actions. In this framework, an agent interacts with an environment and learns to make decisions by receiving feedback in the form of rewards or penalties based on its actions. The core idea is to maximize the cumulative reward over time, which drives the agent to improve its performance through trial and error. This learning paradigm is particularly effective for scenarios where explicit instruction or labeled data is scarce, as it allows the agent to discover optimal behaviors autonomously through exploration.

In contrast, supervised learning involves training a model on a labeled dataset, where the correct output is known in advance, which does not reflect the agent's active learning from environmental feedback. Heterogeneous learning, while a legitimate concept, typically refers to learning from diverse sources or types of data rather than a focus on learning from actions. Transfer learning involves taking a pre-trained model and adapting it to a new task, but it does not specifically address the learning of actions and their consequences in an exploratory manner as reinforcement learning does.