Spanda

Chapter 1 — What is Spanda, anyway?

Back to index

The problem (without the buzzwords)

Building a robot today usually means:

A Python script that calls an LLM
A C++ node that talks to the motor driver
A ROS topic somewhere in the middle
A safety person asking “who stops the robot if the model hallucinates?”
A deploy checklist nobody updates

Nothing is wrong with that stack — it’s just a lot of duct tape.

Spanda is an autonomous systems platform — and at its center is one language (.sd files) where sensors, AI, safety rules, and “will this fit on the Jetson?” live in the same program. The same toolchain also verifies hardware, simulates missions, and monitors health.

The body metaphor

Think of a robot like a body:

Part	In real life	In Spanda
Senses	Eyes, lidar, IMU	`sensor`
Muscles	Wheels, arm, gripper	`actuator`
Reflexes	“Don’t hit the wall”	`safety { }`
Brain	Planner, LLM, vision	`ai_model`, `agent`
Bouncer	Only safe moves get through	`safety.validate()`
Nervous system	The program that ties it together	`behavior`, `task`

Spanda is the pulse — the thing that turns perception, intent, and safety into action.

The word Spanda is Sanskrit for the divine pulse — pronounced SPUN-duh (/ˈspʌndə/). It names the creative vibration that bridges stillness and movement; in software, that is the coordination layer between sensors, AI, safety, and actuators.

What a `.sd` file is

.sd = Spanda source file (think “system definition”)
One file can describe one robot, or a whole fleet
You check it (compiler), run it (simulator), verify it (hardware fit)

You are not writing a web app. You are writing what the robot is allowed to do.

What Spanda is NOT

Myth	Reality
“It’s Python with robots”	It’s its own language with robot keywords built in
“It replaces ROS”	It can talk ROS-style topics; it’s the program layer on top
“The AI drives the robot”	AI proposes; safety approves; actuators only see approved moves
“I need a real robot to try it”	`spanda run` uses a simulated backend

The safety rule everyone remembers

// ❌ Compiler says no
wheels.execute(proposal);

// ✅ This is the pattern
let action = safety.validate(proposal);
wheels.execute(action);

If you remember one thing from this guide: AI output is untrusted until validated.

Try it now

No setup essay — just run an example from the repo:

spanda check examples/basics/01_minimal_robot.sd
spanda run examples/basics/02_sensors_and_safety.sd

If those work, you understand 80% of what Spanda is for.

Next: Your first five minutes

This site is open source. Improve this page.