On this page

Look at

  • New versions of the Agent Look At DOP and Agent Look At Apply DOP.

  • You can now define targets using a point cloud with the new Points target type, or other agents in the crowd as targets with the new Agents target type.

  • You can define the attention span of agents using the Gaze Duration parameter.

  • You can set how interesting a target is to an agent based on its proximity, relative velocity, orientation, and location in the field of view.

  • Look At can now use skeleton adjustment involving the rig’s spine and optional eye joints, with different behavior based on the angle to the target.

  • You can now trigger state transitions based on the current target with the new Look At Target trigger type on the Crowd Trigger DOP.

  • A new version of the Agent Look At SOP applies look-ats at the SOP level, optionally without simulation history.


  • Agent definitions can now store a rest pose for the joints in an agent rig. This is used when loading clips that do not have animation for all joints, and when agents do not have a clip or pose explicitly assigned.

  • Agent from Rig SOP node can now set an agent rig’s rest pose from a point attribute (like rest_transform).

  • Agent SOP now sets the rest pose for an agent when loading it from a subnet, FBX, or USD file.

  • The Agent Unpack SOP's MotionClip output mode now uses the agent rig’s rest pose as the MotionClip’s rest pose instead of leaving its transforms at zero. This improves the behavior of nodes like the Motion Clip Cycle SOP which uses the rest pose.

  • Crowd Solver DOP node can now apply locomotion from layered animation clips (like from the Agent Clip Layer DOP node). It applies locomotion from a layered animation clip when the joint group includes a __locomotion__ joint and the locomotion clip speed is not zero.

  • Agent SOP node now issues a warning when it encounters an invalid skinned USD primitive.

  • Agent Collision Layer SOP now lets you override the bounds scale of each collision shape with the Bound Scale parameter.

  • The Crowd Assign Layers SOP now lets you switch an agent’s assigned layers using one or more search/replace patterns with the new Switch Layers parameter. The search pattern is very similar to the pattern used by the Attrib String Edit SOP.

  • The Agent Terrain Adaptation DOP node now supports layered animation clips when computing the foot locking channel value. You can blend the “foot down” channels from a layered animation clip if its joint group includes the ankle/toe joints.


  • New agentfindclip VEX function looks up an agent clip by name. This is useful when you want to check whether or not an agent clip exists, and can be more efficient when sampling a clip multiple times.

  • A new setagentclips signature sets an agent’s current clips using the indices returned by agentfindclip.

What’s new in Houdini 19.5