EvolvingGrasp: Evolutionary Grasp Generation via Efficient Preference Alignment

TL;DR

This work addresses the generalization gap in dexterous grasping by enabling evolutionary grasp generation that learns from experience through efficient preference alignment. The core methods are Handpose-wise Preference Optimization (HPO), which reframes preference learning as a posterior optimization task, and a Physics-Aware Consistency Model (PCM) that distills a diffusion model into a lightweight, few-step sampler while enforcing physical constraints. The approach pairs rapid inference with iterative preference refinement and physics-guided plausibility, yielding state-of-the-art grasp success rates and sampling efficiency across four benchmarks, including real-world deployment on a Shadow Hand. The results underscore the practical impact of combining diffusion-based preference finetuning with physics-aware consistency distillation for robust, adaptable, and efficient dexterous grasping in diverse environments.

Abstract

Dexterous robotic hands often struggle to generalize effectively in complex environments due to the limitations of models trained on low-diversity data. However, the real world presents an inherently unbounded range of scenarios, making it impractical to account for every possible variation. A natural solution is to enable robots learning from experience in complex environments, an approach akin to evolution, where systems improve through continuous feedback, learning from both failures and successes, and iterating toward optimal performance. Motivated by this, we propose EvolvingGrasp, an evolutionary grasp generation method that continuously enhances grasping performance through efficient preference alignment. Specifically, we introduce Handpose wise Preference Optimization (HPO), which allows the model to continuously align with preferences from both positive and negative feedback while progressively refining its grasping strategies. To further enhance efficiency and reliability during online adjustments, we incorporate a Physics-aware Consistency Model within HPO, which accelerates inference, reduces the number of timesteps needed for preference finetuning, and ensures physical plausibility throughout the process. Extensive experiments across four benchmark datasets demonstrate state of the art performance of our method in grasp success rate and sampling efficiency. Our results validate that EvolvingGrasp enables evolutionary grasp generation, ensuring robust, physically feasible, and preference-aligned grasping in both simulation and real scenarios.

Figures (7)

• Figure 2: Overview of EvolvingGrasp. The evolutionary process begins with the human preference guidance, where Handpose-wise Preference Optimization (HPO, highlighted in the green rectangle) is employed to facilitate preference alignment. These grasp poses are generated by the Physics-Aware Consistency Model (shown in blue rectangles), including Sampling and Distillation mechanism, to ensure the sampling efficiency and the physical plausibility. In this way, EvolvingGrasp, an efficient evolutionary grasp generation framework is proposed to enable the grasp model iteratively converge toward preferred distributions.
• Figure 3: Mean grasping performance in terms of success rate and penetration of randomly selected 6 objects with the finetuning epoch increasing during inference optimization.
• Figure 4: Evolution of robotic grasp preferences during efficient feedback-driven finetuning across 10 epochs. Top row illustrates the adjustment from hand occlusion to clear nozzle visibility. Middle row demonstrates the transition from lens obstruction to an unobstructed camera view. The bottom row shows the evolution from a top-down grasping approach to a bottom-up one, while simultaneously mitigating physical impacts.
• Figure 5: The relationship between the success rate of EvolvingGrasp and the number of generated samples.
• Figure 6: The variation of the mean grasping success rate of the evolving model for randomly selected objects in the multidex dataset with increasing finetuning epochs. The blue dashed line represents the grasping success rate under the original model.