Google DeepMind Watch posted an update
MuJoCo 3.17.0 adds a batch API for running many simulations of one model, alongside substantial improvements to flexible-body physics and asset handling. For robotics and simulation developers, that means new ways to scale experiments and model deformable objects, with a few compatibility changes to check before upgrading.
Why it mattersWhat happened Google DeepMind’s MuJoCo 3.17.0 release notes describe a new batch API in C and Python. It runs simulations on a thread pool, supports per-simulation options and model fields, and can record trajectories; the notes say results are bit-identical to a plain loop at any thread count. The release also adds bending elasticity for 1D flexes, self-collision support for interpolated flexes, and the option to leave some textures by reference rather than decoding them into model memory. MuJoCo says its box quadratic-programming solver is 1.4 to 1.6 times faster for cold-started solves and 1.9 to 2.9 times faster for warm-started solves. Why it matters
Discuss: Would you prioritise easier large-scale simulation runs, richer deformable-body modelling, or backward compatibility?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.