COURSE 27L2100% FREE
Verified 2026-08-14

Sense, plan, act

The classic robotics loop: **sense** the world, **plan** a next move, **act** with actuators, then sense again—under safety constraints. MIT Manipulation’s course description organizes topics as perception, planning (kinematics, collisio...

What it is

The classic robotics loop: sense the world, plan a next move, act with actuators, then sense again—under safety constraints. MIT Manipulation’s course description organizes topics as perception, planning (kinematics, collision-free motion, task-and-motion planning, uncertainty), and dynamics/control (model-based and learning-based). Modern Robotics separates motion planning (Ch. 10) from robot control (Ch. 11).

HIGH PRIORITYDIAGRAM
◷ IN PRODUCTION

Visual Spec & Architecture Diagram

Sense–Plan–Act cycle triangle/loop with icons: cameras/lidar (Sense), planner graph (Plan), motors (Act). Timing note 'classic deliberative loop; modern systems blur layers'.

Educational Focus: Core robotics literacy diagram.

Why it matters

Embodied AI fails differently than chat: physics, latency, and irreversible actions dominate. Gemini Robotics highlights interactivity—replanning when objects slip or move—as a required quality, not a nice-to-have.

How it works (plain)

Sensors → state estimate → planner/policy → controller → motors. Humans supervise high risk. Simulation helps; reality remains the exam. Software frameworks (ROS ecosystem; Isaac Lab for learning loops) structure how modules talk.

Everyday example

Crossing a street: look, decide, step, look again—not a single frozen plan.

Try it

For a robot vacuum, list sensors, plans, acts, and one required human override.

Myths

⚠️ Myth: A smart VLM replaces the control stack.
✓ Reality: DeepMind’s Gemini Robotics-ER story still interfaces with low-level safety-critical controllers per embodiment.
⚠️ Myth: Slow planning is always safer.
✓ Reality: Too-slow loops miss dynamics—design for latency budgets (supervisory-control chapter).

Sources