Lagrange v1.1.0: easier to use, checked against measurements (release notes)

v1.1 is the first feature release on the stable 1.x line. It makes Lagrange easier to start and to drive: one lagrange command, one run file, robots added while it runs. It adds plugins, walking people, real robot models, water, a wind tunnel and rotor downwash, a realistic viewport, and many new sensors. Several models are now compared with measured data. The documentation is published as a site with a physics reference for every model.

Everything frozen by RFC 0007 changed only by addition. The plugin ABI went from 1.0 to 1.1 by appending, and new schemas were added (CustomSensorAPI_1, HumanAPI_1, WaterAPI_1, ThermalCameraAPI_1, JointDynamicsAPI_1, ElectricalBranchAPI_1 and others), all checked by the frozen-surface tests. v1.0 files, plugins, profiles and recordings still load.

Status: released on 2026-11-07 as v1.1.0, on the owner’s request. Documentation: https://open-physical-systems.github.io/Lagrange/.

Highlights

Easier to run

  • lagrange runs the server and the viewer together, the way gazebo does (lagrange amr, lagrange quad, lagrange sedan, lagrange swerve, lagrange town, or lagrange ROBOT [WORLD]). It also runs a profile with the viewer (lagrange PROFILE.yaml).

  • The run file (docs/guide/run-file.md) holds a whole run in one YAML file: robot, world, what to simulate and the site. Your own robot starts still on an empty ground.

  • Spawning. Start without a robot, then add and remove robots and people while it runs: with lagrange spawn and lagrange delete, File → Add robot in the viewer, or ROS 2 simulation_interfaces.

  • Joint control. Any robot can be driven joint by joint with no controller, over /joint_commands.

  • Site changes while running. lagrange set gravity=1.62 wind_speed=5 and the viewer’s Environment panel change the site as it runs: gravity, wind, temperature, rain, fog, pressure and humidity. The panel also has weather presets.

  • lagrange check says which joints cannot hold the robot. lagrange bake splits a URDF’s concave collision meshes into convex parts.

  • Examples (examples/): one ROS 2 package per use, each run by a launch file. They cover your own robot, an arm, a power system, a lidar, fog, the quadcopter, the Holybro X500 V2 by teleop, a car, spawning, ros2_control, a C plugin, learning, a trotting quadruped, a camera and the wind tunnel.

  • Python. lagrange.envs offers Gymnasium’s interface over a simulation, with environments running in parallel.

  • A TX16S-style remote on screen (utils/lagrange_tx16s). Fly it from a real RadioMaster TX16S over USB or from the keyboard.

Plugins and people

  • Plugins in the server (RFC 0008). Plugins are loaded with --plugin, through plugin ABI 1.1. A plugin can:

    • see the world and the scene (rays, overlaps);

    • simulate custom sensors (the sample lagrange_tof time-of-flight and ultrasonic sensor);

    • map signals to and from ROS topics (--ros-topics);

    • replace the physics engine, which a conformance suite then checks.

    Start from the plugin template.

  • Walking humans (RFC 0009). People are 14 kinematic capsule segments. They walk with their stance foot planted and steer round robots and each other. Add them as crowds (--humans N), one by one (lagrange spawn) or as SDFormat actors. A thermal camera sees them at skin temperature.

Robots and worlds

  • Real models, fetched with checksums (fetch_assets):

    • the Lexus RX 450h (Autoware’s sample vehicle);

    • the Holybro X500 V2 (PX4’s frame);

    • the AMR_FMS swerve AMR in its 60 m × 40 m warehouse.

  • Real places built from open data, Zurich and Nishi-Shinjuku in Tokyo.

  • The town. Working lights, zebra crossings whose buttons people press, vans and buses, lane changes, and four seasons with snow and ice.

  • Water (WaterAPI_1, --water): buoyancy, drag, added mass, currents and linear waves. Fossen’s model takes a vehicle’s identified coefficients, and the BlueROV2 Heavy’s are included.

  • A trotting quadruped example.

Physics

  • Rotor downwash. The jet under a drone, its outwash on the ground, and the wake against walls and roofs, felt by other robots and checked against OpenFOAM hover cases. A live lattice Boltzmann air box can run round the drone on the GPU (downwash_model: lbm), and the wind blows through it.

  • Rotors: the vortex-ring state from NeuroBEM’s fit, the ceiling effect, and the air’s torque turning the propeller so a multirotor yaws.

  • The wind tunnel (lagrange tunnel): lattice Boltzmann on the collision shapes, on the GPU or the deterministic CPU reference, with a grid study. lagrange foam writes and runs the same case in OpenFOAM.

  • The weather reaches the drone. Rotors, drag and the controller fly in the weather’s air density, and rain and snow push links.

  • Electrical:

    • velocity and voltage motor modes, and motors on sliding joints;

    • a SPICE netlist or a URDF <lagrange> block as the power system;

    • resistors, fuses, switches and diodes between buses (RFC 0010).

  • Rigid body:

    • joint damping, friction and mimic joints from every robot format;

    • each body’s contact force, and a force-torque reading at each joint, in MuJoCo and Jolt;

    • mesh colliders as convex hulls;

    • gravity changed while it runs.

Sensors

  • Lidar intensity from surface materials, incidence angle and range.

  • Lidar motion distortion on both tiers.

  • Sensor latency and jitter, applied the same way for ROS, plugins, recordings, Python and the viewer.

  • A thermal camera (ThermalCameraAPI_1): Planck band radiance, with a Swinbank clear sky.

  • Each camera’s instance image on ROS 2.

  • Camera sensors drawn with the viewport’s realistic renderer.

The viewer

  • Realistic shading, the default on a GPU: sky, sun shadows, physically based materials, real water, photo textures, and falling rain and snow.

  • The Electrical workspace gains Circuit, Netlist and 3D overlay tabs, and Auto-layout.

  • A battery’s State cards and OCV plot.

  • The Signals list gains a unit filter and drag onto the scope.

  • The Environment panel gains live site entry and weather presets.

  • An angle flight mode for multicopters. Airspeed and tilt are shown beside the ground speed.

Checked against measurements

tools/ci/realism.sh runs these checks (docs/validation/report.md):

Check

Reference

Result

Battery

LG HG2 cell (Kollmeyer et al.): pulses and drive cycles

5.4–6.2 mV RMS from 46 to 89 % SoC. 21.5 mV at 24 % misses the 20 mV target. Unseen LA92 11.7 mV, US06 20.1 mV (just over). No temperature dependence

Planar pushing

MCube (Yu et al. 2016)

MuJoCo within 6 mm and 4–7° of the quasi-static model. Jolt does not turn the block

Rotors in flight

NeuroBEM flights (Bauersfeld et al. 2021)

coefficients 1.46 N RMS (their fit 1.49). bem 1.12 N (their BEM 0.98)

Drone under a drone

Kiran et al. 2025

76 % of thrust kept at 4 l against 65 % measured (11 points over)

Underwater vehicle

BlueROV2 Heavy (von Benzon et al. 2022)

Top speeds 0.729 / 0.626 / 0.713 m/s against 0.72 / 0.63 / 0.71

Wind tunnel

Ahmed body (SAE 840300)

Over four times too much drag. Wall friction 3.1 times too high

Documentation

  • A documentation site (Sphinx, published to GitHub Pages from main). It has diagrams, videos and highlighted code.

  • An installation guide, from ROS 2 Lyrical to a first run.

  • A physics reference for every model: what it computes, its equations, how it is solved, and how far to trust it.

  • A Robots and worlds page showing each sample running, with its command.

  • New tutorials: your own drone step by step, a power system from a schematic, people around your robot. There are also feature videos of real runs.

Known issues

  • The misses in the table above, and everything listed as owed in the validation report. There is still no sim-vs-real recording of a real robot (the owner has no hardware to record).

  • Not run while GitHub Actions are disabled: the sanitizer builds and the nightly jobs. The release was checked by a local colcon build and colcon test --packages-above lagrange_core.

  • Radiation stays a zero substitute.

Release checklist

  • [x] Release notes

  • [x] Full local build and test

  • [x] Documentation site built with warnings as errors, and published

  • [x] Released: dev merged into main, tagged v1.1.0, release/1.x brought up to it

  • [ ] Packages published and signed (maintainers, packaging/README.md)