RELEASE.md
This release brings recent performance benefits that where CUDA only to all GPU backends while relaxing CUDA Toolkit requirement. Besides, all reported CUDA crashes have been fixed.
This release continues on the ongoing trend of rigid body simulation speed improvements. A few camera-related bugs and all known regression on Metal backend are now fixed.
The numerical stability of the simulation for simple rigid objects has been greatly improved. Apart from that, rigid body simulation is now much faster for complex scenes with many entities. Finally, a significant number of bugs have been fixed.
This release introduces more sensors while significantly speeding up collision detection on GPU (up to 30%). As usual, a few bugs have been fixed.
This release introduces a new type of tactile sensors based on FOTS, and add support of parallel simulation of heterogeneous articulated robots. Beyond that, the performance of the simulation has been significantly improved, especially when rigid option 'noslip' is enabled.
This release mainly improves experimental IPC coupler integration. Avatar entity has been re-introduced after being broken for months. Finally, Quadrants' dynamic arrays are now support on Apple Metal to avoid systematic scene compilation.
This release polishes our newly introduced USD parser and external interactive viewer plugin mechanism. Beyond that, the performance of the simulation has been significantly improved for collision-heavy scenes (up to 30%) and robots using capsule/sphere collision geometries (up to 20%).
This release mainly focuses on usability, by extending support of USD and introducing a new external plugin mechanism for the interactive viewer. Besides, the performance of the simulation has been significantly improved for collision-heavy scenes (up to 30%).
This small release adds user-friendly diagnosis of invalid Rigid physics properties and improves support of GLTF meshes.
This PR focuses on performance improvements (x4 faster for complex scenes with 64 < n_dofs < 96 and n_envs=4096 compared to 0.3.10). Besides, initial support of heterogenous object and USD stage import for rigid body simulation has been introduced.
The main focus of this release is to improve scaling of the simulation wrt the complexity of the scene, and better leverage GPUn compute for small to moderate batch sizes (0<=n_envs<=8192). As usual, a bunch of minor bugs have been fixed.
Small release mainly fixing bugs.
Small release mainly polishing features that were introduced in previous release.
The performance of data accessors have been dramatically improved by leveraging zero-copy memory sharing between GsTaichi and Torch. Beyond that, the robustness of the default contact algorithm has been improved, and differentiable forward dynamics for Rigid Body simulation is not partially available. Last, but not least, GsTaichi dynamic array mode is finally enabled back by default!
The performance of GsTaichi dynamic array mode has been greatly improved. Now it should be on par with fixed-size array mode (aka performance mode) for very large batch sizes, and up to 30% slower for non-batched simulations. This mode is still considered experimental and must be enabled manually by setting the env var 'GS_ENABLE_NDARRAY=1'. Just try it if you are tired of endlessly waiting for the simulation to compile!
diffuse_texture to Glass surface. (@Kashu7100) (#1934)A new experimental interface with the Incremental Potential Contact coupling solver libuipc has been introduced, mainly targeting cloth simulation.
Minor release mainly aiming at polishing existing features and addressing some major performance regression that was introduced end of august. GsTaichi dynamic array type and fast cache are now enabled by default on Linux and Windows (opt-in on MacOS via env variable 'GS_ENABLE_NDARRAY'), which should help avoiding recompilation in most cases.
This minor release mainly introduces first-class sensor support (IMU, Contact Sensor, LiDAR, Depth camera and more), incl. recording and plotting facilities. The rigid-rigid hydroelastic contact model has also been added. As usual, a fair share of bugs have been fixed, with unit test coverage gradually improving.
This minor release fixes a few non-blocking rendering issues for the Rasterizer backend.
This minor release fixes a few additional regressions and initiates migration to our own open-source fork of Taichi, GsTaichi (contributions are welcome!).
This small release addresses the most pressing regressions that has been pointed out by the community since 0.3.0. Support of coupling between Rigid Body and FEM has been improved and should be more reliable, though it is still considered experimental for now. Apart from that, no behavior changes are to be expected.
This release focuses primarily on stability, covering everything from MJCF/URDF parsing to rendering and physics, backend by a new CI infrastructure running more than 200 unit tests on all supported platforms. The most requested Mujoco features that were previously missing have been implemented. Native support of batching has been extended to all solvers except Stable Fluid, motion planning, and rendering via gs-madrona. Finally, support of soft body dynamics has been enhanced, with the introduction of constraints and coupling between soft and rigid body dynamics.
We would also like to acknowledge the ongoing PRs. Some of them have not yet been merged because we have not had enough time to fully test them: