The new Agent Add ML Deformer SOP configures agent shapes to be deformed using a model trained with the ML Train Deformer recipe for higher-quality skin deformation. This supports real-time deformation in the viewport for large crowds.

Version 2.0 of the Agent Configure Joints SOP provides improved interactive configuration of joint limits, matching KineFX’s Configure Joints SOP.

Version 2.0 of the Agent Collision Layer SOP supports automatically generating collision shapes from skin geometry, matching KineFX’s Ragdoll Collision Shapes SOP.

Enhanced round-tripping between KineFX skeletons and agents, with new parameters for rest transforms, joint limits, and joint groups in the Agent Unpack SOP, Agent Animation Unpack SOP, Agent Character Unpack SOP, and Agent from Rig SOP.

Version 2.0 of the Agent Constraint Network SOP and Test Simulation: Ragdoll SOP support the new joint limit format in the Agent Configure Joints SOP v2.0.

The Crowd MotionPath Edit SOP’s viewer state now provides a transform handle for manipulating the selected pinned points (press K to toggle the transform handle). The SOP is also more robust in handling changes to the input motion paths.

Improvements to the Crowd MotionPath Follow SOP include more robust handling of guide curves that self-intersect and smoother steering when enforcing the distance variance limit.

The Crowd MotionPath Layer SOP now supports randomly varying the Blend Ratio parameter or overriding it with an attribute.

On the Crowd State DOP, the Time Dependent parameter specifies whether the random clip selection should vary over time. This produces different random clip selections if agents transition back into the same state more than once in a simulation.

On the Agent SOP and Agent Clip SOP, you can now control the clip’s Frame Range when importing from USD. In addition to the previous behavior of using the stage’s time code range, the frame range can also be determined from the time samples of attributes on the SkelAnimation and Skeleton primitives, or be overridden with a custom frame range.

The Agent SOP’s USD importer now includes channels for all the blendshapes defined on skinned primitives. This allows the channels to be available on the agent regardless of whether the SkelAnimation primitive contains animation for the channels. Previously, only blendshape channels defined on the SkelAnimation primitive were included.

On the Agent Layer SOP, the Remove Shapes and Remove Layers parameters remove existing shapes and layers from the agent definition. These parameters are also exposed on the SOP Import Crowd LOP for pruning shapes and layers that are not needed when exporting to USD.

On the Agent Transform Group SOP, the Leaf Transforms parameter specifies a pattern of transform(s) that end the traversal from the Root Transforms. This can be useful for selecting a limited set of joints instead of the entire subtree under the root transforms.

The Agent Terrain Adaptation SOP now exposes the Sampling Method parameter for terrain projection, which was previously only available in DOPs.

On the Agent Terrain Adaptation DOP, when leaning is turned on (Lean Angle Per Frame) and the agent has multiple sets of legs (for example, a quadruped), the terrain normal is sampled under each “lower back” joint. This produces better results when the agent is moving over bumpy terrain. Previously, each joint used the terrain normal under the agent’s point.

On the Crowd Procedural LOP, the Bake Skinning for Prototype Agents parameter controls whether deformed geometry is generated for the prototype agents.

On the SOP Import Crowd LOP, the Subframe Sampling parameters and Cache Current Frame Only mode are now exposed from the Cache LOP for easier configuration of samples for motion blur.