FixTheRobot.dev · interactive robotics labs

Learn Robotics by Solving Real Engineering Problems

Fix broken robots in your browser. Find out why a controller oscillates, a robot crashes or a map drifts, repair it, and get scored on randomized hidden tests. Free, with nothing to install.

Start Robot Foundations Try a live demo Log in to save progress

All tracks and demos

Try autonomous navigation Try the humanoid Try Physical AI Beat the planner

Explore Robotics

Ten learning tracks. New to robotics? Start with Robot Foundations. Control and Sensors have three labs each, in a suggested order.

Suggested path: Robot Foundations Control and Sensors Perception Localization & SLAM Navigation Robot Learning (in development). Linear Algebra for Robotics is the maths companion to all of them. Physical AI and Vision-Language-Action build on all of these.

Control labs

New to control? Take the three labs in order. The first two need no coding.

Control · Step 1 of 3 Tune a PID Controller Beginner Move three sliders and learn what Kp, Ki and Kd really do to a robot lift. About 10–15 minutes · No coding or robotics background. You move sliders. Control · Step 2 of 3 Fix the Oscillating Robot Beginner A mobile robot zig-zags around its path. Read the plots, tune Kp and Kd, and prove your fix survives disturbances. About 15–20 minutes · Recommended first: Tune a PID Controller. No coding. Control · Step 3 of 3 Build a CBF Safety Filter Intermediate Write a control barrier function in Python that keeps a mobile robot from crashing, graded on 20 random scenarios. About 30–45 minutes · Basic Python (functions, if/else, arithmetic). Recommended first: the two Control labs before it.

Coming next in Control: LQR stabilization and MPC trajectory tracking.

Sensors labs

Point clouds: from raw LiDAR numbers to obstacles, in three labs. Take them in order.

Point clouds · Step 1 of 3 Understand & Visualize a Point Cloud Beginner Your robot just received its first LiDAR scan. Inspect the XYZ points in Python, convert range and angles to coordinates, and explore the scene in 3D. About 30–40 minutes · Basic Python (lists, loops, functions). NumPy is not needed. Point clouds · Step 2 of 3 Process & Transform a Point Cloud Beginner Clean a raw frame (NaN, Inf, zero returns), filter by range and region, move it from the sensor frame to the robot frame and downsample it with a voxel grid. About 40–50 minutes · Needs the first point-cloud lab: Understand & Visualize a Point Cloud. Point clouds · Step 3 of 3 Build a LiDAR Obstacle Detection Pipeline Intermediate From raw frame to STOP or GO: remove the ground, cluster obstacles, measure their distance and check whether the robot's path is clear. About 45–60 minutes · Needs the second point-cloud lab: Process & Transform a Point Cloud.

More Sensors labs are on the way.

Try it live

Four interactive simulations of how real robots work. Each one loads only when you open it.

Autonomous navigation Click a destination and watch the robot sense, plan and drive there. SimulationLaunch
Humanoid motion Tell a 3D humanoid to move an object. It walks, reaches and grasps. 3D simulationLaunch
Physical AI Give a robot a task in plain language. It plans, acts, checks and recovers. SimulationLaunch
Beat the planner Race a path planner through an unknown map. Can you find the shorter route? GameLaunch

Humanoid Motion Playground

Tell a humanoid what to do.

Watch it stand, walk, reach, grasp, carry, and move objects. Orbit, zoom and pan to inspect the robot from any angle.

Simulation. A kinematic, educational model of a research-style humanoid. No real robot is connected, and the motion is generated live (nothing is a pre-recorded animation).

State: STANDING
Balance: STABLE
Support: DOUBLE FOOT
COM: CENTERED

Loading the 3D simulation…

World #1 · Seed: 0

Physical AI / humanoid stack

TaskYour command
Task plannerSteps and skills
World modelWhat it believes
PerceptionFind the object
Motion plannerPath to walk
Footstep plannerWhere each foot goes
IKJoint angles for the arm and legs
BalanceKeep the COM over the feet
Whole-body motionLegs, torso and arms together
RobotMoves

Task execution log

Nothing yet.

Structured system state, not a transcript of thoughts.

Physical AI Playground

Tell the robot what to do.

Give a robot a task in natural language. Watch it understand, plan, perceive, navigate, manipulate, verify, and adapt.

Simulated environment. No physical robot is connected, and nothing you type leaves your browser.

World #1 Seed: 0

Moving obstacle

Ready. Pick an example or type a command.

LanguageYour command
Task plannerSteps and skills
World modelWhat it believes
PerceptionCamera, lidar
NavigationPath and avoidance
ManipulationArm and gripper
VerificationDid it work?
ReplanAdapt and retry

World state

Perception

Navigation

Manipulation

Verification

Task execution log

Nothing yet.

Structured system state, not a transcript of thoughts: this is what the robot's software holds right now.

Game

Can you beat the planner?

Every world is different. You don't know the map, and neither does the robot.
Take turns moving one step at a time.
Reach the goal using as few steps as possible.

Challenge #1 Seed: 0

YOUR TURNMove one step

You

Steps 0

Planner

Steps 0

Arrow keys or WASD also work.

Waiting for you to move.

Ready?

You can only see what your robot discovers.

Take turns moving one step at a time. Reach the goal using as few steps as possible.

Not discovered yet
Open floor
Wall
Goal
Your path
Planner's path
Shortest possible

Robot navigation playground

Make the robot go anywhere

Click a destination. Watch it figure out the rest.

Ready. Click the map to send the robot there.

👆 Click anywhere to send the robot there

Want to see how the robot is doing this?

Every trip goes through the same loop. The step that is working right now lights up.

Your goalWhere you clicked
SensorsA spinning range scanner
World awarenessWhat is where, and what is moving
Path planningThe best route on the map
Obstacle avoidanceSlow, stop or re-plan
ControlSteering and speed
RobotMoves

Waiting for a goal.

How did the robot do that?

It did not follow a recording. Four parts of robotics worked together, live, in your browser.

Plan
Find a path to the destination that goes around walls and shelves.
Sense
Detect obstacles around the robot, including people and robots that move.
Localize
Understand where the robot is on the map, so the plan means something.
Control
Turn the plan into smooth steering, speed and stopping.

Learn how robots actually do this

Robot FoundationsStart here: how a robot senses, moves and plans ControlSteering, PID and safe motion SensorsLiDAR and point clouds PerceptionFrom pixels and points to objects, tracks and scenesOpen Localization & SLAMWhere am I, and what does the map look like?Open Large Language ModelsBuild a GPT from scratch, then fine-tune and evaluate itOpen NavigationPlanning and avoiding, like the demo aboveOpen Systems EngineeringSelect, integrate, debug and deploy a whole robotOpen UAV Systems EngineeringFrom UAV fundamentals to GNSS-denied autonomous flightOpen

Research Landscape Twenty areas of robotics, from Physical AI to SLAM: how each evolved, with the key papers, people and open problems. Explore the research → News The day's most important robotics stories, ranked by importance and how many outlets report them. Read the news → Robotics Updates New papers, open-source releases and industry news, collected automatically from public sources. See what's new →

How a lab works

Every lab is one real engineering problem, a simulator, and an automatic grader. You learn the theory at the moment you need it.

ProblemA robot misbehaves: it crashes, oscillates or drifts.
ExperimentRun the simulation and watch exactly how it fails.
CodeWrite the controller in Python, right in the page.
EvaluateHidden randomized tests score safety, performance and effort.
UnderstandTargeted feedback, three levels of hints, and a worked solution.

New

Robotics Research Landscape

A living map of the top 20 areas of robotics, from Physical AI and humanoids to SLAM, control and soft robots. Real papers, researchers, labs, code and open problems, plus a side-by-side survey of all twenty.

Explore the Research Landscape Start anywhere: each field has a timeline from foundations to the frontier, essential reading and a research radar.

Read first

Control barrier functions explained with a mobile robotWhat h(x) means, why the CBF condition keeps a robot safe, and how a safety filter works. CBF vs potential fields for robot obstacle avoidanceTwo ways to avoid obstacles, and why only one of them can guarantee safety. Why is my robot oscillating? A practical guide to controller tuningHow to read the error plot and fix oscillation with Kp and Kd.

Learn Robotics by Solving Real Engineering Problems

Explore Robotics

Control labs

Sensors labs

Try it live

Tell a humanoid what to do.

Physical AI / humanoid stack

Task execution log

Task complete

What just happened?

Tell the robot what to do.

Task understanding

World state

Perception

Navigation

Manipulation

Verification

Task execution log

Task complete

You just used a Physical AI system

Can you beat the planner?

You

Planner

Challenge complete

You just experienced online path planning.

Make the robot go anywhere

✓ Destination reached!

How did the robot do that?

Plan

Sense

Localize

Control

Learn how robots actually do this

How a lab works

Robotics Research Landscape

Latest Robotics

Read first