Lagrange v1.3.0: 37 new sensors, a faster simulator, hardened protocol (release notes)¶
v1.3.0 is a feature release on the stable 1.x line. Everything is an addition (RFC 0007): existing robots, worlds, profiles, plugins and recordings run as before.
Status: released on 2026-11-14 as v1.3.0, on the owner’s request. Documentation:
https://open-physical-systems.github.io/Lagrange/.
Install¶
curl -fsSL https://raw.githubusercontent.com/Open-Physical-Systems/Lagrange/main/packaging/install.sh | bash
source /opt/ros/lyrical/setup.bash
lagrange amr
Sensors¶
37 built-in sensor kinds (a URDF
<sensor type="KIND">or USDCustomSensorAPI_1, no plugin): power (battery monitor, INA226/ACS712 current and voltage, ESC telemetry, tether load cell), environment and drones (anemometer, airspeed, rangefinder, light, sun, rain, humidity, visibility, optical flow, particles, ADS-B), AMR and arms (bumper, cliff, wheel odometry with real slip, safety scanner with protective and warning fields, force/torque, joint torque, proximity, gripper, line sensor, RFID/QR tags, UWB, Wi-Fi), cars (vehicle CAN, TPMS, V2X), underwater (sonar), cameras (event camera, stereo/RGB-D, detections with 2D/3D boxes and semantic segmentation, landing target), and a 4D imaging radar. Each cites its datasheet or reference; every default was checked against its source (docs/plan/sensor-defaults-verification-2026-11-12.md). Guide:docs/guide/sensors.md; examplelagrange_example_sensors.On ROS 2 by default on standard messages (Range, LaserScan, Odometry, BatteryState, WrenchStamped, vision_msgs Detection2D/3DArray, …), images with
camera_info.Sensor faults for every sensor: dropout, stuck, bias, drift, noise, disconnect (
--sensor-fault, run file, profile, Python).Lidar scan patterns (Velodyne channel table, flash, Livox rosette) and fisheye cameras up to 360°, in URDF, SDFormat and USD (RFC 0015).
Python
sim.image()returns any image signal as a numpy array.
Speed¶
Run (real-time factor, headless) |
v1.2.1 |
v1.3.0 |
|---|---|---|
Sedan in the town |
0.8 |
3.7 |
Town, no robot |
3.7 |
about 22 |
Sedan, city block |
1.1 |
12.9 |
Quadcopter |
14.5 |
about 60 |
AMR, warehouse |
7.3 |
31.8 |
The camera model and the thermal camera run on the GPU, the camera render overlaps the rest
of the step, the radar runs in parallel (and on the GPU with --lidar gpu), MuJoCo skips far
kinematic bodies, ROS messages are built only for subscribers, and the viewer draws only what
changed (loading a scene 1.85 s → 37 ms). Start-up and spawning are faster too. Measured on a
shared laptop, ±30 %: docs/design/benchmarks.md §4–§5.
Security and fixes¶
Protocol limits before the handshake (RFC 0014): 64 KiB frames until the client is greeted, connection limits, a handshake timeout (
--max-connections,--max-pending,--handshake-timeout,--peer-rate). Before, any client could make the server hold 64 MiB per connection, with no limit on connections.Two QA passes (
docs/plan/qa-2026-11-12.md,docs/plan/qa-sensors-2026-11-12.md) fixed about 56 problems, among them a hugeStepfreezing the server, 15 and 30 Hz lidars and IMUs refused at start, spawned robots’ sensors reading the first robot’s parts, a second multicopter grounding the first, images lost onrestore_state, NaN accepted from the command line and Python.
Upgrading from 1.2.x¶
Saved states (
Server::saveState, Pythonsave_state()) now use format 2 and keep camera images; a state saved by 1.2.x is refused. Recordings are not affected.IMU
<dynamic_bias_stddev>is now read as Gazebo reads it (the driving noise of the bias).Some sensor noise defaults changed to their datasheet values (SHT40, WindSonic, BH1750, VL53L0X, SPS30).
A server now refuses more than 32 connections (16 not yet greeted) by default.
The quadcopter run is 3–5 % slower than before the third perf round (owed).
Checked¶
Whole workspace (66 packages) built and tested on Ubuntu 26.04 / ROS 2 Lyrical: 1799 tests,
0 failures; lagrange determinism identical for the AMR, sedan and quadcopter profiles.