BusyBox: Benchmarking Affordance Generalization

BusyBox is a physical 3D-printable device for benchmarking affordance generalization in robot foundation models.

It features

• Modular design with 6 interchangeable modules (buttons, switches, sliders, wires, knob, and display)
• Open-source CAD files and bill of materials for easy reproduction
• Optional electronics and Raspberry Pi instrumentation for automated state logging
• Reconfigurable setups enabling systematic evaluation of generalization
• A language-annotated dataset of 1000+ demonstration trajectories oof BusyBox affordances

Please check out our website for more details.

For fully building a instrumented BusyBox capable of state logging, see the BOM.

First print the BusyBox following Printing Instructions with details on files to print and any details on print settings.

TODO: add instructions on how to assemble electronics with pictures

Instructions for flashing the Arduino Nano's firmware: Flashing Firmware

See Data Collection for details on our data collection methodology.

The robots/aloha/ directory contains a complete data recording pipeline for collecting teleoperated demonstrations using the Aloha bimanual robot.

• Aloha robot hardware (leader + follower arms) set up and calibrated
• USB foot pedal for episode control
• (Optional) Instrumented BusyBox with MQTT bridge for state logging

pip install -r requirements.txt
python robots/aloha/record_busybox_episodes.py

Edit robots/aloha/utils/config.py to set:

• dataset_dir — where episodes are saved
• camera_names — which cameras to record
• using_instrumented_busybox — enable MQTT state logging

Each episode is saved as an HDF5 file with:

• /observations/qpos — joint positions (14,) per timestep
• /observations/qvel — joint velocities (14,)
• /observations/effort — joint torques (14,)
• /observations/images/{cam_name} — JPEG-compressed camera frames
• /action — leader arm joint commands (14,)
• /observations/timestamp, /action_timestamp — timing data

For instrumented BusyBox setups, the devices/pi_sw/ directory contains the MQTT bridge that runs on the Raspberry Pi:

# On the Raspberry Pi
pip install pyserial paho-mqtt
python devices/pi_sw/mqtt_bridge.py --broker-host <MQTT_BROKER_IP>

See Flashing Firmware: Installing MQTT on the Pi for setup details.