A gentle intro to Robotics
Before you build a robot, it helps to know the words. This tutorial is a short tour of the ideas the rest of the course uses, one chapter at a time, each with a figure you can play with. Nothing here needs hardware.
Two problems: locomotion and manipulation
Most of robotics is one of two problems.
Locomotion is moving the robot itself, on wheels, tracks, or legs. It answers the question how do I get there? A warehouse cart driving to a shelf, a four-legged robot climbing stairs, and a vacuum mapping a living room are all locomotion. Getting to a goal takes three steps:
- Path planning builds the plan: it finds a route from where the robot is to the goal that avoids every obstacle.
- Trajectory optimization turns that route into a trajectory the robot can follow, smooth and within its speed limits, saying where it should be at each moment.
- Control drives the motors so the robot follows that trajectory, correcting it whenever it drifts.
Together with knowing where the robot is along the way, these steps are called navigation.
Locomotion: a wheeled robot drives itself around two obstacles to a goal. The dashed line is the planned path; the solid line is the trajectory it has followed so far.
Manipulation is changing the world with a hand. Grasping, placing, pushing, turning, using a tool. It answers the question what do I do with this object? Picking a cup off a table, folding a shirt, and tightening a screw are all manipulation.
Manipulation: an arm reaches a cup, closes its fingers on it, carries it across the table, and sets it down.
Some robots do both: an arm on a wheeled base, or on a legged body. That is a mobile manipulator.
Why Solarian starts with manipulation
Locomotion is largely solved. Legged robots trained with reinforcement learning in simulation walk over rough ground, recover from a push, and climb stairs. Wheeled robots build a map, locate themselves on it, and plan a route through a building. You can buy both today.
Manipulation is still a work in progress, and it is the hardest open problem in robotics. Every object has its own shape, weight, and friction. The hand has to touch the world, and contact is hard to predict. A grasp a few millimetres off slips. A task that looks trivial to a person, such as loading a dishwasher, hides hundreds of these small decisions.
So Solarian focuses on manipulation first: an arm, a gripper, a camera, and the tools to teach them a task from demonstrations.
The robot in this tutorial
Loading MuJoCo model…
The robot this tutorial builds: the SO-106 arm, the UMI Slim gripper on its flange, and the camera on the gripper, looking along the fingers. Each label is the module's name in the project.
Every part of this tutorial uses the same robot: the SO-106, a six-joint desktop arm built from Feetech servos, with a UMI-slim two-finger gripper and a camera on the wrist. You build it in simulation first, so you can follow the whole course before you own any parts.
The chapters
This first tutorial is organized into five chapters:
- The robot's body. Links, joints, end effectors, and degrees of freedom, with the SO-106 as the example.
- Robot configurations and coordinates. How to write down where the robot is, as joint angles or as a hand pose, and the frames those numbers live in.
- Kinematics. Converting between joint angles and hand poses, and the space the hand can reach.
- The control loop. Sensors and what they return, actuators and torque, and the control that moves a joint along a trajectory.
- Policies and learning. What a policy is, the language of states, actions, and trajectories, and how policies are learned from demonstrations and from reward: diffusion, ACT, VLAs, and reinforcement learning.
Already know the basics? Skip ahead to Your first Solarian robot and come back when a word shows up. The glossary is the longer list.
Next: the robot's body.