BulletLab

Developed by Ranasurya Ghosh

A robotics experimentation framework that transforms PyBullet robots into intuitive Python objects, with modern ImGui-based controls, telemetry, visualization, and reinforcement learning workflows.

Install BulletLab library: pip install bulletlab

 

---

What is BulletLab?

BulletLab provides a high-level object-oriented interface to PyBullet that simplifies robotics experimentation by exposing joints, links, sensors, and environments as intuitive Python objects instead of raw physics engine IDs. It combines real-time simulation with a ImGui-powered modern interface for interactive control, parameter tuning, telemetry visualization, and experiment management, while also offering reinforcement learning integration for training and evaluating autonomous robotic systems within a unified workflow.

Instead of this:

p.setJointMotorControl2(
    robot_id, joint_index,
    controlMode=p.VELOCITY_CONTROL,
    targetVelocity=15,
    force=100
)

You write this:

robot.joints["motor"].velocity = 15

BulletLab Arsenal: The Official Package Registry

Just as Python has PyPI for software packages, BulletLab has Arsenal for verified robotics assets.

Arsenal is the official registry of the BulletLab ecosystem. It solves the long-standing problem of hunting down compatible URDFs and manually fixing missing meshes. Arsenal provides:

• Verified Robot Packages: Curated, community-contributed models guaranteed to load correctly.
• One-Line Installation: Permanently download assets to your local machine (Robot.install()).
• Direct Loading: Stream assets directly into your session cache without permanently modifying your filesystem (Robot.load("arsenal:...")).
• Standardized Package Format: Powered by machine-readable manifests (metadata.json) for automated validation.

Whether you are conducting reproducible research or building quick demos, Arsenal ensures you spend less time configuring assets and more time writing robotics code.

---

Architecture

BulletLab uses a two-window architecture:

Window	Purpose
PyBullet Native Window	Physics simulation, 3D rendering, camera
BulletLab ImGui Window	Control panels, telemetry, live plots, console

These windows communicate through Python objects. BulletLab does not attempt to replace PyBullet's renderer or embed ImGui inside the simulation viewport.

---

Quick Start

Installation

Install from PyPI

pip install bulletlab

Developer Installation

git clone https://github.com/NuclearVenom/BulletLab.git
cd BulletLab
pip install -e .

Basic Example

from bulletlab import Simulation, Robot
from bulletlab.ui import BulletLabUI

# Create simulation
sim = Simulation()
sim.start()

# Load robot
robot = Robot.load("path/to/robot.urdf", sim=sim)

# Control joints by name
robot.joints["wheel_left"].velocity = 10
robot.joints["wheel_right"].velocity = 10

# Modify physics parameters
robot.links["chassis"].mass = 5.0
robot.links["wheel_fl"].friction = 1.2

# Get robot state
state = robot.get_state()
print(f"Position: {robot.base_position}")
print(f"Roll: {robot.roll:.2f}°")

# Build UI
ui = BulletLabUI(sim=sim)
ui.register_panel(...)
ui.run()

Telemetry & Logging

from bulletlab.telemetry import TelemetryManager
from bulletlab.logging import DataLogger

telemetry = TelemetryManager()
telemetry.watch("Speed", lambda: robot.base_velocity[0])
telemetry.watch("Roll",  lambda: robot.roll)

logger = DataLogger()
logger.watch("speed", lambda: robot.base_velocity[0])
logger.start("run1.csv")

for _ in range(1000):
    sim.step()
    telemetry.update()
    logger.step()

logger.stop()

Live Plotting

from bulletlab.plotting import LivePlot

plot = LivePlot(title="Robot Speed")
plot.watch("Speed", lambda: robot.base_velocity[0], color="#00ff88")
plot.start()

for _ in range(1000):
    sim.step()
    plot.update()

Camera Follow

from bulletlab import Simulation, Robot, CameraFollow

sim = Simulation(mode="gui").start()
robot = Robot.load("husky/husky.urdf", sim=sim, position=(0, 0, 0.3))

# One line — camera glides after the robot (smooth mode by default)
cam = CameraFollow(robot, sim)

# Or pick a mode:
cam = CameraFollow(robot, sim, mode="snap")    # locks instantly
cam = CameraFollow(robot, sim, mode="smooth")  # cinematic glide
cam = CameraFollow(robot, sim, mode="chase")   # always behind the robot

while sim.is_connected:
    sim.step()
    cam.update()   # ← one call keeps the camera centred on the robot

Hover Highlighting

from bulletlab import Simulation, Robot, RobotHighlighter
from bulletlab.ui import BulletLabUI

sim = Simulation(mode="gui").start()
robot = Robot.load("kuka_iiwa/model.urdf", sim=sim)

# One line — hover any joint/link in the UI to see it glow in 3D
hl = RobotHighlighter(robot, sim)
app = BulletLabUI(sim=sim, robots=[robot], highlighter=hl)
app.run()

Hovering over an Explorer row or a Properties slider instantly highlights the matching 3D part in the PyBullet window with an orange pulsing glow.

BulletLab Arsenal

BulletLab Arsenal is the official robot asset registry — load community robots with a single line, no manual download required.

from bulletlab import Simulation, Robot
from bulletlab.core.world import World

sim = Simulation(mode="gui").start()
World(sim).load_plane()

# Install permanently to ~/.bulletlab/packages/
Robot.install("reference_bot")

# Load directly from Arsenal into a session cache (cleaned up on exit)
robot = Robot.load("arsenal:reference_bot", sim=sim, position=(0, 0, 0.3))

# Load a specific model
robot = Robot.load("arsenal:reference_bot/BLem1", sim=sim)

All standard Robot.load() parameters — position, fixed_base, tilt, etc. — work identically with Arsenal sources.

ImGui Control Panel

from bulletlab.ui import BulletLabUI
from bulletlab.ui import widgets as ui

app = BulletLabUI(sim=sim, robots=[robot])

@app.custom_panel("My Controls")
def my_panel():
    ui.button("Reset", robot.reset)
    ui.slider("Wheel Mass", robot.links["wheel"].mass, 0.1, 20,
              setter=lambda v: setattr(robot.links["wheel"], "mass", v))
    ui.checkbox("Motors Enabled", lambda: motors_on,
                setter=lambda v: toggle_motors(v))

app.run()

Virtual Joystick

Add an interactive 2D joystick to any custom panel for intuitive, gamepad-style robot control.

@app.custom_panel("Drive")
def drive_panel():
    ui.joystick(
        "Rover Drive",
        on_y=lambda v: [setattr(robot.joints["wheel_left"],  "velocity", v * 10),
                        setattr(robot.joints["wheel_right"], "velocity", v * 10)],
        on_x=lambda v: [setattr(robot.joints["wheel_left"],  "velocity", -v * 5),
                        setattr(robot.joints["wheel_right"], "velocity",  v * 5)],
        snap=True,                       # Handle snaps to center on release
        size=80,                         # Joystick circle radius (pixels)
        handle_color=(0.2, 0.8, 0.4, 1.0),
    )

• snap=True — robot stops when you release (default)
• snap=False — handle latches; robot keeps driving at the last commanded speed
• Drop multiple joysticks in the same panel (use ui.same_line() between them)
• Each joystick can have its own color and size

---

Supported Robot Types

BulletLab is completely generic — no code assumes a specific robot type:

• Cars & rovers
• Drones & quadrotors
• Robotic arms
• Self-balancing robots
• Quadrupeds
• Humanoids
• Custom mechanisms

---

Reinforcement Learning

BulletLab exposes clean state/action interfaces without depending on any ML framework:

# Compatible with any RL approach
state = robot.get_state()      # → numpy array
action = my_policy(state)      # → numpy array
robot.apply_action(action)     # → updates joints

# Manual Q-learning, SARSA, evolutionary algorithms — all supported

---

Examples

Example	Description
examples/01_differential_drive_rover.py	Rover with wheel velocity control
examples/02_robotic_arm.py	Joint position control with ImGui sliders
examples/03_self_balancing_robot.py	PD controller for balance
examples/04_drone_parameter_tuning.py	Thrust/mass parameter exploration
examples/05_generic_robot_inspector.py	Load any URDF and inspect it
...and more

Run any example:

python examples/01_differential_drive_rover.py

---

Documentation

The full documentation is available online at https://nuclearvenom.github.io/BulletLab/

Building Locally (For Contributors)

If you wish to edit or preview the documentation locally:

pip install -e ".[dev]"
mkdocs serve

Then visit http://localhost:8000

---

Testing

pip install -e ".[dev]"
pytest tests/ -v --cov=bulletlab --cov-report=term-missing

---

Technology Stack

Component	Library
Physics	PyBullet
UI	Dear ImGui (pyimgui)
Data	NumPy, Pandas
Config	PyYAML
Plotting	PyQtGraph
Testing	PyTest
Docs	MkDocs + mkdocstrings

For AI Agents & LLMs

BulletLab is designed to be highly predictable and LLM-friendly. If you are an AI agent writing code for a user:

1. Read llms.txt in the repository root for a dense, AI-optimized API summary.
2. Check the Cookbook & Snippets for copy-pasteable implementations of common tasks.
3. Use the robot.joints[name] API over pybullet integer IDs whenever possible.

---

Community

We welcome contributions and feedback! Check out our community resources:

• Contributing Guide – How to build, test, and contribute to BulletLab
• Code of Conduct – Our community standards
• Security Policy – How to report security vulnerabilities responsibly
• Roadmap – Our vision for future releases
• Citation – How to cite BulletLab in academic research

---

License

MIT License — see LICENSE for details.