Research map / Probabilistic and scientific modeling / Dynamical systems
JODA: Composable Joint Dynamics for Articulated Objects
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes JODA, which represents articulated-joint dynamics as a three-channel PCHIP field (conservative force, dry friction, damping) composed from effect templates proposed by a VLM and refined via editing, feedback, and differentiable MJX simulation.
Problem
Articulated assets used in simulation lack fine-grained joint-level dynamics (friction holding, detents, soft closing, snap latching), and existing approaches either ignore such structure or use overly simple spring/damping models.
Approach
Represent joint dynamics as a composable, interpretable three-channel PCHIP field over normalized joint coordinate; a VLM proposes semantic effect templates (type, interval, qualitative strength) that are compiled into physically scaled fields, with closed-loop diagnostic feedback and differentiable MJX refinement.
Key results
- Gradient-based refinement reduced cabinet trajectory MSE from 4.10e-2 to 1.56e-3 after 9 Adam iterations.
- JODA takes 0.062 ms per simulation step on average, 1.37x the original MuJoCo cost.
- On the refrigerator/kitchen cabinet quasi-static force comparison, JODA more closely matches measured curves than a linear spring baseline (RMSE reported in figure, values not given in text).
- Across gravity scenes, JODA satisfies the initial effect specification in 2.2 proposal rounds on average.
Surprising findings
- A linear spring cannot produce nonzero force at equilibrium, limiting its match to measured nonlinear opening forces.
- Removing the template library and asking the VLM for raw control points significantly degrades reproducibility.
Limitations
- Single DOF with three position-dependent channels; no multi-joint couplings, hidden-state, history-dependent or nonlinear damping.
- Dry friction and joint-limit constraints are non-smooth and need local smoothing for differentiability.
- Validation is largely qualitative with few objects and a small quantitative comparison.
Experimental setup
Datasets: PartNet-Mobility Lightwheel Sim-Ready Assets real-world cabinet trajectories measured quasi-static opening forces from prior work
Benchmarks: PartNet-Mobility Lightwheel Sim-Ready Assets
Metrics: RMSE vs measured quasi-static force curves trajectory MSE simulation step time overhead VLM proposal token counts qualitative trajectory comparison
Baselines: constant friction/damping model linear spring return model no-template ablation (direct VLM control-point output)
Builds on: MuJoCo MJX PCHIP
Closest papers by meaning
- AdaManip: Adaptive Articulated Object Manipulation Environments and Policy Learning (ICLR 2025 · similarity 0.87)
- CoDA: Coordinated Diffusion Noise Optimization for Whole-Body Manipulation of Articulated Objects (NeurIPS 2025 · similarity 0.86)
- GMAP: Generalized Manipulation of Articulated Objects in Robotic Using Pre-trained Model (AAAI 2025 · similarity 0.86)
- ChainHOI: Joint-based Kinematic Chain Modeling for Human-Object Interaction Generation (CVPR 2025 · similarity 0.86)
- DIPO: Dual-State Images Controlled Articulated Object Generation Powered by Diverse Data (NeurIPS 2025 · similarity 0.86)
- URDF-Anything+: End-to-End Generation for Simulation-Ready Articulated Assets (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.