Robotics loops in small, readable Python

PythonInteractiveRobotics

Robots observe, act, fail, retry, update beliefs, and replan. This gallery shows those loops with runnable examples and generated GIFs, using only numpy and matplotlib in core.

Star on GitHub Try live Open in Colab Run the first loop See the examples
39 runnable examples
38 generated GIFs
111 smoke and regression tests
0 ROS, GPU, or simulator required

Run the first loop

The first example is a tiny tabletop robot that misses a grasp, updates its belief, and retries differently. 

git clone https://github.com/rsasaki0109/PythonInteractiveRobotics.git
cd PythonInteractiveRobotics
python3 -m pip install -e .
python3 examples/manipulation/01_pick_and_retry.py

Run in Colab

These notebooks clone the repo, install the lightweight package, run the real example headless, print the trace summary, and show the generated GIF.

Pick and retry Safety filter CBF Human correction replanning Clarifying question Household task agent

Closed-loop robotics examples

Each GIF is generated from the same runnable example linked below. The point is the feedback loop: observe, update state, act, and recover when the world disagrees.

A tabletop robot misses grasps, updates belief, and retries.

Pick and retry

grasp miss -> belief update -> retry

Open example
A point robot's nominal velocity is projected onto a CBF safe set.

Runtime safety filter

nominal controller -> CBF projection -> safe motion

Open example
An active SLAM agent reduces pose and map uncertainty.

Active SLAM toy

pose belief -> map update -> information action

Open example
A two-link arm predicts an occluded moving target and reaches it.

Reaching under occlusion

observe target -> predict through occlusion -> correct

Open example
A grid robot observes a hidden wall and replans with A star.

Online replanning

map update -> invalidate path -> A* replanning

Open example
A grid robot receives human correction, raises a zone's cost, and replans around it.

Human correction replanning

shortcut -> human feedback -> cost update -> replan

Open example
A toy VLA loop parses a goal, reads visual tokens, fails, and retries.

Tiny VLA loop

language goal -> visual tokens -> skill retry

Open example
A tabletop robot asks which block before picking from an ambiguous command.

Clarifying question

ambiguous command -> ask -> answer -> act

Open example
A household robot clarifies an ambiguous command, replans around unsafe and human-corrected zones, retries a grasp, and stores the block.

Household task agent

clarify -> plan -> safety check -> retry -> human replan

Open example
A sorter asks for help when the conformal prediction set is ambiguous.

Conformal ask-for-help

calibrate -> prediction set -> place or ask

Open example
A grid robot plans through high-empowerment cells.

Empowerment navigation

reachable successors -> shaped A* -> open route

Open example
A grid robot learns a reward from one demonstration and plans to a new goal.

Inverse reward from demo

demo features -> learned weights -> shaped planning

Open example

Choose a path through the repo

The GitHub repo keeps the full example index, contributor guide, trace contract, and bridge plans for ROS2 and simulators.

Repository README Shareable live trace Complete example index Learning paths Trace contract Example authoring guide Public launch checklist