Skip to content

Changelog: v2.16 ​

v2.16.0 (2026-11-15) ​

RaiSim

  • Runtime joints. World::addRuntimeJoint() connects two rigid objects or articulated links, also of different robots, with a Fixed, Ball, Hinge or Slider joint between simulation steps, and World::removeRuntimeJoint() removes it again. The joints keep the objects separate; their impulses are solved together with contacts, tendons and joint limits. The world XML (<runtime_joint>) saves them with their enabled and broken state, and checkpoints keep the state of the joints that exist; adding or removing a joint makes earlier checkpoints unrestorable. See Runtime joints. The new runtime_joint_benchmark times them.
  • World::discardPendingStep() cancels a step prepared by integrate1() or integrateNoContactDetection() without solving, integrating or advancing the world time. It clears the prepared contacts and constraint rows and allows adding and removing objects, tendons and wires again; preparation already done (waking objects, advancing projectiles) is not rolled back.
  • Articulated systems use less memory and reuse their dynamics scratch storage between steps. MatDyn copies no longer allocate for empty matrices.
  • Windows: the native FEM mesher build compiles in RaiSim's message logger, and its external projects use explicit build directories.
  • The FEM mesher reads and tessellates more STEP files; six complex parts (a DC motor, a hollow section, a pneumatic cylinder, a servo horn, a socket screw and a sprocket) are checked against independent Open CASCADE measurements, and a knee mount is a browser regression case.
  • The raisim_engine2 world editor (the raisim_engine2/ tree and the RAISIM_ENGINE2 CMake option) was removed from this repository.

raisimPy

  • raisimpy.RuntimeJoint and World.addRuntimeJoint(), getRuntimeJoint(), getRuntimeJoints() and removeRuntimeJoint(); the runtimeJoints.py example.
  • World.discardPendingStep().

Examples and documentation

  • The C++ runtime_joints example.
  • The examples build with MSVC's /bigobj.
  • The Newton benchmark pages compare ten-step control intervals and random actuation, and normalize performance by hardware price.

RaiSim — physics for robotics and AI.