Despite transformative advances in generative motion synthesis, real-time interactive motion control remains dominated by traditional techniques. In this work, we identify two key challenges in bridging research and production: 1) Real-time scalability : Industry applications demand real-time generation of a vast repertoire of motion skills, while generative methods exhibit significant degradation in quality and scalability under real-time computation constraints, and 2) Integration : Industry applications demand fine-grained multi-modal control involving velocity commands, style selection, and precise keyframes, a need largely unmet by existing text- or tag-driven models. Moreover, a systematic motion design interface for generative models remains absent. To overcome these limitations, we introduce MotionBricks: a large-scale, real-time generative framework with a two-fold solution. First, we propose a large-scale modular latent generative backbone tailored for robust real-time motion generation, effectively modeling a dataset of over 350,000 motion clips with a single model. Second, we introduce smart primitives that provide a unified, robust, and intuitive interface for authoring both navigation and object interaction. Notably, MotionBricks applies to new downstream tasks in a zero-shot manner, where no fine-tuning or task-specific tagging is required. Applications can be designed in a plug-and-play manner like assembling bricks without expert animation knowledge, enabling an accessible interface for applications in animation and robotics. Quantitatively, we show that MotionBricks produces state-of-the-art motion quality on open-source and proprietary datasets of various scales, while also achieving a real-time throughput of 15,000 FPS with 2ms latency. We demonstrate the flexibility and robustness of MotionBricks in a complete production-level animation demo, covering navigation and object-scene interaction across various styles with a unified model. To showcase our framework's application beyond animation, we deploy MotionBricks on the Unitree G1 humanoid robot to demonstrate its flexibility and generalization for real-time robotic control.

Abstract We present a general method for transferring skeletons and skinning weights between characters with distinct mesh topologies. Our pipeline takes as inputs a source character rig (consisting of a mesh, a transformation hierarchy of joints, and skinning weights) and a target character mesh. From these inputs, we compute joint locations and orientations that embed the source skeleton in the target mesh, as well as skinning weights to bind the target geometry to the new skeleton. Our method consists of two key steps. We first compute the geometric correspondence between source and target meshes using a semi‐automatic method relying on a set of markers. The resulting geometric correspondence is then used to formulate attribute transfer as an energy minimization and filtering problem. We demonstrate our approach on a variety of source and target bipedal characters, varying in mesh topology and morphology. Several examples demonstrate that the target characters behave well when animated with either forward or inverse kinematics. Via these examples, we show that our method preserves subtle artistic variations; spatial relationships between geometry and joints, as well as skinning weight details, are accurately maintained. Our proposed pipeline opens up many exciting possibilities to quickly animate novel characters by reusing existing production assets.

We propose a fully automatic method for specifying influence weights for closed-form skinning methods, such as linear blend or dual quaternion skinning. Our method is designed to work with production meshes that may contain non-manifold geometry, be non-watertight, have intersecting triangles, or be comprised of multiple connected components. Starting from a character rest pose mesh and skeleton hierarchy, we first voxelize the input geometry. The resulting sparse voxelization is then used to calculate binding weights, based on the geodesic distance between each voxel lying on a skeleton "bone" and all non-exterior voxels. This yields smooth weights at interactive rates, without time-constants, iteration parameters, or costly optimization at bind or pose time. By decoupling weight assignment from distance computation we make it possible to modify weights interactively, at pose time, without additional pre-processing or computation. This allows artists to assess impact of weight selection in the context in which they are used.