ML-based robotics refers to robotic systems that rely on machine learning techniques instead of explicitly programmed rules. Rather than encoding behavior through hand-crafted logic, these systems learn from data, demonstrations, or interaction with an environment. Common approaches include supervised learning from expert trajectories, reinforcement learning through trial and error, and unsupervised representation learning. ML-based robotics enables robots to handle complex sensory inputs such as vision and language, adapt to new tasks, and generalize beyond predefined scenarios. This paradigm contrasts with traditional robotics, which depends on precise models and control equations. While ML-based methods often require large datasets and careful training, they offer greater flexibility and scalability, especially in unstructured or dynamic environments.