MuJoCo 3.15.0 adds experimental soft-body elasticity, improves flex-body mechanics and corrects how actuator damping is calculated. The release also changes an API in ways that could alter damping behaviour in existing simulations, making this more than a quiet patch for robotics and physics developers.
Google DeepMind Watch analysis
What happened
Google DeepMind’s MuJoCo 3.15.0 release notes, dated 5 October, describe changes to the open-source physics engine. The update spans simulation mechanics, collision handling, sensors and model-file support.
Our top picks
- More accurate actuator damping
The calculation now accounts for tendon and actuator armature, and inertial coupling. Some multi-link setups may have lower damping than before. - Experimental Stable Neo-Hookean elasticity
A new option supports simplified elasticity for non-interpolated 3D flexes. It is experimental and requires the discrete integrator. - Better flex-body mechanics
Flex bending and stretching now include motion and reaction forces from articulated vertex attachments. - Capsule multicontact
Single-shot multicontact collision handling has been added for capsule geometries. - A new clearance sensor measurement
The insidesite sensor can report signed clearance or protrusion, measuring how much an object juts out of a site. - Support for larger model buffers
Binary model save and load functions now accept buffer sizes above 2 GiB.
Why it matters
For people building robotics and physics simulations, the changes reach into how soft bodies deform, how contacts are handled and how actuators respond. The revised damping calculation is especially worth checking: the release notes say multi-link actuators may exhibit lower damping, while tendon- or site-driven actuators spanning multiple joints should be damped more accurately.
Our read
This is a useful engineering release, not a reason to rebuild every simulation before lunch. The experimental elasticity option and the changed damping behaviour deserve targeted tests, while the sensor and large-model improvements are practical additions for the right workloads. Check the release notes and validate any affected actuator setups before relying on old tuning assumptions.
What to watch
- How the new damping calculation behaves in existing multi-link and tendon-driven models.
- Whether the experimental elasticity option proves useful beyond early testing.
- Follow-up changes to flex attachments, collision handling and sensor workflows.
Discussion spark: For robotics simulations, would you prioritise more physically accurate actuator damping even if it changes the behaviour of existing models, or favour backwards-compatible results?
Sources and evidence
- 3.15.0 (5 October 2026, 14:58 UTC)
not affiliated with, endorsed by, or operated by Google or Google DeepMind