IK-GAT Kinematic Graph Attention Architecture
Figure 2: IK-GAT architecture. The model accepts 3D joint positions as input, which are projected and augmented with learnable positional embeddings. The core encoder leverages a Kinematic Prior , defined by the skeletal parent-child relationships, to structure the Graph Attention (GAT) layers. The network employs a dual-residual design, featuring both local skip connections within GAT blocks and a global shortcut from the input projection. The head regresses rotation matrices representing bone-aligned world frames via a 6D parameterization.
Paper context
Paper title: Amortized Inverse Kinematics via Graph Attention for Real-Time Human Avatar Animation Abstract: Inverse kinematics (IK) is a core operation in animation, robotics, and biomechanics: given Cartesian constraints, recover joint rotations under a known kinematic tree. In many real-time human avatar pipelines, the available signal per frame is a sparse set of tracked 3D joint positions, whereas animation systems require joint orientations to drive skinning. Recovering full orientations from positions is underconstrained, most notably because twist about bone axes is ambiguous, and classical IK solvers typically rely on iterative optimization that can be slow and sensitive to noisy inputs. We introduce IK-GAT, a lightweight graph-attention network that reconstructs full-body joint orientations from 3D joint positions in a single forward pass. The model performs message passing over the skeletal parent-child graph to exploit kinematic structure during rotation inference. To simplify learning, IK-GAT predicts rotations in a bone-aligned world-frame representation anchored to rest-pose bone frames. This parameterization makes the twist axis explicit and is exactly invertible to standard parent-relative local rotations given the kinematic tree and rest pose. The network uses a continuous 6D rotation representation and is trained with a geodesic loss on SO(3) together with an optional forward-kinematics consistency regularizer. IK-GAT produces animation-ready local rotations that can directly drive a rigged avatar or be converted to pose parameters of SMPL-like body models for rea Passages referencing this figure: the inverse task of recovering a set of local rotations given 3D joints, such that FK ( { 𝐑 t , loc ( i ) } i = 0 N − 1 ) ≈ 𝐉 ~ t \mathrm{FK}(\{\mathbf{R}^{(i)}_{t,\mathrm{loc}}\}_{i=0}^{N-1})\approx\tilde{\mathbf{J}}_{t} . This is ill-posed: joint positions underconstrain rotations, most notably through twist around bone-aligned axes and in the presence of noisy or missing joints, as shown in Fig. 1 . Key idea. We separate the IK problem into a learned component and a deterministic component. Rather than regressing local rotations directly, we learn an intermediate representation that (i) is stable for regression and (ii) admits an exact, closed-form conversion to standard local rotations. The network predicts per-joint rotation matrices representing bone-aligned world frames , while t network predicts per-joint rotation matrices representing bone-aligned world frames , while the conversion to animation-compatible parent-relative local rotations is analytic and lossless. This reduces the complexity of what must be learned: the model is not required to learn kinematic composition rules or rest-pose basis conventions, only the pose-dependent orientations in a canonicalized frame. Figure 1: Twist ambiguity problem. Identical joint positions (blue dots) are produced by any rotation around the bone axis (red). This makes the inverse problem ill-posed without strong anatomical priors. Neural IK methods attempt to learn these priors from data. The bone-aligned world rotation matrices differ from the standard world-space rotation matrices in how the underlying local coordinate f
