cs.ROApr 2, 2026

O-ConNet: Geometry-Aware End-to-End Inference of Over-Constrained Spatial Mechanisms

Authors: Haoyu Sun, Meng Zhao, Tianhao Wang, Jianxu Wu

Organizations: School of Mechanical, Electronic and Control Engineering, Beijing Jiaotong University, Beijing, China.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Deep learning has shown strong potential for scientific discovery, but its ability to model macroscopic rigid-body kinematic constraints remains underexplored. We study this problem on spatial over-constrained mechanisms and propose O-ConNet, an end-to-end framework that infers mechanism structural parameters from only three sparse reachable points while reconstructing the full motion trajectory, without explicitly solving constraint equations during inference. On a self-constructed Bennett 4R dataset of 42,860 valid samples, O-ConNet achieves Param-MAE 0.276 +/- 0.077 and Traj-MAE 0.145 +/- 0.018 (mean +/- std over 10 runs), outperforming the strongest sequence baseline (LSTM-Seq2Seq) by 65.1 percent and 88.2 percent, respectively. These results suggest that end-to-end learning can capture closed-loop geometric structure and provide a practical route for inverse design of spatial over-constrained mechanisms under extremely sparse observations.

Figures & tables

Fig. 1 : D-H parameterization schematic of the Bennett mechanism.
Fig. 2 : 3D trajectory visualization of dataset samples. Each panel displays the complete motion trajectory of a Bennett mechanism’s third joint endpoint, with speed magnitude encoded by a plasma colormap.
Fig. 3 : Complete architecture of O-ConNet. On the left, a permutation-invariant kinematic state encoder processes 3 sparse observation points through a shared-weight MLP and learnable attention pooling (AttnPool), compressing them into a global latent vector z∈R512\mathbf{z}\in\mathbb{R}^{512} . In the center, a Transformer decoder expands z\mathbf{z} into 64-point trajectory coordinates T^\hat{\mathbf{T}} and a velocity field V^\hat{\mathbf{V}} . On the right, the parameter prediction head adopts a ResNet-style shortcut-residual dual-path structure, where the shortcut branch maps z\mathbf{z} directly to coarse logits s\mathbf{s} (supervised by Laux\mathcal{L}_{\mathrm{aux}} ) and the residual branch reads the reconstructed trajectory to produce a zero-initialized refinement r\mathbf{r} ; the merged output is decoded via Sigmoid and atan2 into (a^12,α^12)(\hat{a}_{12},\hat{\alpha}_{12}) . In all trajectory visualizations throughout this paper, blue solid lines denote ground truth (GT), red dashed lines show the network prediction (Pred), green dotted lines represent the forward kinematic reconstruction from predicted parameters (Kin), and orange stars mark the input observation points.
Fig. 4 : Trajectory prediction and kinematic verification comparison for three representative validation samples. Each row shows one sample; from left to right: 3D view and XY/XZ/YZ projections. Color coding follows Fig. 3 .
Fig. 5 : 3D trajectory comparison of ablation configurations and baselines across randomly selected validation samples. Color coding follows Fig. 3 . Baselines without a trajectory-decoding branch (MLP-Direct, PointNet-Only) predict parameters directly and thus show no Pred curve (red dashed).
 Sub-groupMethodParamsParam-MAE ↓\downarrowa12\mathbf{a_{12}} -MAE ↓\downarrowα12\boldsymbol{\alpha_{12}} -MAE ↓\downarrowTraj-MAE ↓\downarrowVel-MAE ↓\downarrow
Proposed O-ConNet (Full)5.00M0.276±0.0770.276{\pm}0.0770.0063±0.0013\mathbf{0.0063{\pm}0.0013}0.316±0.063\mathbf{0.316{\pm}0.063}0.145±0.018\mathbf{0.145{\pm}0.018}0.017±0.002\mathbf{0.017{\pm}0.002}
AblationEnc. PoolingMaxPool+AttnPool5.13M0.260±0.084\mathbf{0.260{\pm}0.084}0.0065±0.00080.0065{\pm}0.00080.337±0.1180.337{\pm}0.1180.149±0.0250.149{\pm}0.0250.017±0.0030.017{\pm}0.003
MaxPool only4.97M0.283±0.1090.283{\pm}0.1090.0067±0.00100.0067{\pm}0.00100.339±0.1310.339{\pm}0.1310.175±0.0310.175{\pm}0.0310.018±0.0030.018{\pm}0.003
MaxPool+AvgPool5.10M0.286±0.1140.286{\pm}0.1140.0065±0.00120.0065{\pm}0.00120.336±0.1220.336{\pm}0.1220.154±0.0330.154{\pm}0.0330.018±0.0040.018{\pm}0.004
Param. Headw/o ResNet Residual4.97M0.391±0.0790.391{\pm}0.0790.0088±0.00180.0088{\pm}0.00180.502±0.1060.502{\pm}0.1060.199±0.0170.199{\pm}0.0170.021±0.0010.021{\pm}0.001
w/o Aux. Branch Loss5.00M0.301±0.0760.301{\pm}0.0760.0056±0.00090.0056{\pm}0.00090.345±0.0900.345{\pm}0.0900.160±0.0250.160{\pm}0.0250.018±0.0020.018{\pm}0.002
TABLE I : Comprehensive performance summary on the validation set. MLP-Direct and PointNet-Only lack trajectory decoding branches ( † ); corresponding metrics are marked “–”. Each ablation alters only one factor while keeping all others consistent with the full model. All metrics are computed in the normalized coordinate system; lower is better. All values are mean ±\pm std over 10 independent runs.
More in this topic

Trajectory Reconstruction

See all 9 papers

Explore similar work

CardsList

  1. Flexible-body Modeling, Kinematic Identification, and Assembly Accuracy of Overconstrained Spatial Linkages

    Sep 15, 2026Daniel Huczala, Michael Pieber, Johannes Gerstmayr +6Mechanism DesignMultibody Dynamics

    Department of Mechanical Engineering, Ulsan National Institute of Science and Technology (UNIST), 50 UNIST-gil, Eonyang-eup, Ulju-gun, Ulsan, 44919, South Korea · Department of Machine Parts and Mechanisms, VSB – Technical University of Ostrava, 17. listopadu 2172/15, Ostrava, 70800, Czech Republic · Department of Mechatronics, University of Innsbruck, Technikerstraße 13, Innsbruck, 6020, Austria +2

  2. LOInK: Learned Optimal Inverse Kinematics via Structured Neural Surrogate Models

    Sep 18, 2026Michael Somerfield, Damian Abood, Ruigang Wang +1Inverse KinematicsRobot Kinematics

    Australian Robotic Inspection and Asset Management (ARIAM) Hub, the Australian Centre for Robotics (ACFR), and the School of Aerospace, Mechanical and Mechatronic Engineering, The University of Sydney, Australia

  3. RAM: Reachability Across Morphologies

    Jun 8, 2026Tim Walter, Xinyu Chen, Jonathan Külz +1Reachability AnalysisImplicit Neural Representations

    Department of Computer Engineering, Technical University Munich, Germany · German Electron Synchrotron, University of Hamburg, Germany