[Paper Review] GraspADMM: Improving Dexterous Grasp Synthesis via ADMM Optimization

Liangwang Ruan, Jiayi Chen+2 more|arXiv (Cornell University)|Mar 14, 2026

Robot Manipulation and Learning0 citations

TL;DR

GraspADMM uses an ADMM-based refinement to decouple target object contact points from actual hand contacts, achieving state-of-the-art grasp success with diverse, collision-free, and dynamically stable grasps. It improves over Dexonomy by optimizing dynamic stability while strictly enforcing kinematic feasibility.

ABSTRACT

Synthesizing high-quality dexterous grasps is a fundamental challenge in robot manipulation, requiring adherence to diversity, kinematic feasibility (valid hand-object contact without penetration), and dynamic stability (secure multi-contact forces). The recent framework Dexonomy successfully ensures broad grasp diversity through dense sampling and improves kinematic feasibility via a simulator-based refinement method that excels at resolving exact collisions. However, its reliance on fixed contact points restricts the hand's reachability and prevents the optimization of grasp metrics for dynamic stability. Conversely, purely gradient-based optimizers can maximize dynamic stability but rely on simplified contact approximations that inevitably cause physical penetrations. To bridge this gap, we propose GraspADMM, a novel grasp synthesis framework that preserves sampling-based diversity while improving kinematic feasibility and dynamic stability. By formulating the refinement stage using the Alternating Direction Method of Multipliers (ADMM), we decouple the target contact points on the object from the actual contact locations on the hand. This decomposition allows the pipeline to alternate between updating the target object points to directly maximize dynamic grasp metrics, and adjusting the hand pose to physically reach these targets while strictly respecting collision boundaries. Extensive experiments demonstrate that GraspADMM significantly outperforms state-of-the-art baselines, achieving a nearly 15\% absolute improvement in grasp success rate for type-unaware synthesis and roughly a 100\% relative improvement in type-aware synthesis. Furthermore, our approach maintains robust, physically plausible grasp generation even under extreme low-friction conditions.

Motivation & Objective

• Address the four core challenges of dexterous grasp synthesis: diversity, kinematic feasibility (no penetrations), dynamic stability (secure contact forces), and speed.
• Leverage Dexonomy’s sampling/diversity and simulation-based collision resolution while introducing mobility in contact points.
• Formulate refinement as ADMM to decouple target object contact points from actual hand contact points.
• Optimize dynamic grasp metrics while ensuring strict collision-free hand-object contact through auxiliary variables and alternating updates.

Proposed method

• Maintain Dexonomy’s four-stage pipeline (template annotation, sampling-based initialization, simulator-based refinement, template bootstrapping).
• Replace Dexonomy’s purely kinematic refinement with an ADMM-based optimization that decouples object contact points from hand contact points.
• Define grasp quality as a differentiable metric e that accounts for force-closure stability via a QP-based contact force balance.
• Solve a three-step ADMM loop: (1) update target object contact points p^o to maximize e; (2) update hand pose q and hand contact points p^h via forward simulation to satisfy p^h ≈ p^o; (3) update dual variable λ with annealing to stabilize convergence.
• Use forward physics simulation (MuJoCo) with transposed Jacobian control to map contact forces to joint torques, ensuring exact collision handling.
• Adopt a moderate penalty parameter ρ (≈ 10^3) to balance dynamic stability and kinematic feasibility, avoiding pure optimization of either objective.

Figure 1 : Overview . From the same initialization, our GraspADMM framework generates robust grasps by optimizing contact points on the hand (red) and the object (yellow), while Dexonomy [ 6 ] simply fixes these points.

Experimental results

Research questions

• RQ1 How can we improve kinematic feasibility and dynamic stability simultaneously for dexterous grasp synthesis without sacrificing diversity?
• RQ2 Can decoupling target object contact points from actual hand contacts via ADMM enable effective optimization of grasp metrics while maintaining collision-free hand-object interactions?
• RQ3 Does ADMM-based refinement outperform purely simulator-based or purely gradient-based refinements across diverse grasp types and friction conditions?
• RQ4 What is the impact of initialization and hyperparameters (ρ, dual variable annealing) on convergence and grasp quality?

Key findings

Method	GSR %	OSR %	CDC	PD	Div %
DexGraspNet [32]	12.1	57.0	7.6	4.9	29.0
FRoGGeR [19]	10.3	55.7	5.0	0.2	27.0
SpringGrasp [8]	7.8	35.4	23.6	16.6	70.2
BODex [7]	49.2	96.6	3.0	0.6	32.5
Dexonomy [6]	60.5	96.5	0.21	0.0	34.2
GraspADMM (Ours)	74.6	97.2	0.17	0.0	25.7

• GraspADMM achieves a nearly 15% absolute improvement in grasp success rate for type-unaware synthesis over Dexonomy on a large object set.
• GraspADMM delivers roughly a 100% relative improvement in grasp success rate for type-aware synthesis across diverse grasp types.
• The method maintains zero penetration depth and comparable speed to Dexonomy while offering higher diversity due to flexible contact point optimization.
• Under extreme low-friction conditions (μ=0.1), GraspADMM still generates diverse and stable multi-fingered grasps, outperforming baselines.
• Ablation shows that initialization and intermediate ρ values (e.g., 10^2–10^3) are critical for high GSR/OSR; too large ρ degrades performance.

Figure 2 : ADMM Optimization Pipeline . (a) Update target object contact points $\mathbf{p}^{o}$ via gradient descent to maximize dynamic stability. (b) Update hand pose $\mathbf{q}$ and points $\mathbf{p}^{h}$ via forward-simulated transposed Jacobian control to satisfy kinematic feasibility. (c) U