Robotics and AI
Robotics and AI intersect in the creation of autonomous physical agents that perceive their environment, reason about it, and take actions in the physical ...
Intelligent Physical Agents
Robotics and AI intersect in the creation of autonomous physical agents that perceive their environment, reason about it, and take actions in the physical world. A robot integrates hardware including sensors, actuators, and mechanical structure, with software for perception, reasoning, planning, and control. AI provides the intelligence that makes robots adaptable to unstructured environments rather than restricted to executing fixed preprogrammed movements.
The core challenges unique to physical robots that pure software AI does not face include real-time constraints, since decisions must be made fast enough for physical safety; mechanical uncertainty, where motors may not execute commands exactly as specified; and the sim-to-real gap, where behaviours learned in simulation may fail when deployed on physical hardware due to differences in friction, lighting, and sensor noise.
Modern autonomous robots typically employ a layered architecture. The reactive layer handles immediate sensor-to-actuator responses, such as obstacle avoidance, at millisecond timescales. The deliberative layer performs planning and reasoning at second-to-minute timescales. The executive layer bridges the two, decomposing high-level plans into sequences of reactive behaviours. This three-layer architecture, associated with the Sense-Plan-Act paradigm, has been refined and adapted across decades of robotics research.
Simultaneous Localisation and Mapping, or SLAM, is one of the most important AI problems in robotics. A robot exploring an unknown environment must build a map of its surroundings while simultaneously estimating its own position within that map. These two problems are interdependent: accurate mapping requires knowing your position, and accurate localisation requires a map.
Learning in Robotics
Deep reinforcement learning has enabled robots to learn complex manipulation tasks such as grasping novel objects, opening doors, and folding clothes from raw sensory input without hand-coded controllers. OpenAI's Dactyl system learned to solve a Rubik's cube with a human-like robotic hand entirely through simulated self-play reinforcement learning.
Interview Tip
Robotics interviews often ask about the difference between kinematics and dynamics. Kinematics describes the geometry of motion, specifying positions and velocities of robot links as functions of joint angles, without considering forces. Dynamics adds mass, inertia, and forces to predict how the robot will move under given torques. Knowing that inverse kinematics solves for joint angles needed to place the end effector at a desired position is a key foundational concept.