Simulation-Driven Evolutionary Motion Parameterization for Contact-Rich Granular Scooping with a Soft Conical Robotic Hand

Yongliang Wang; Cristian C. Beltran-Hernandez; Tomoya Takahashi; Masashi Hamaya

cs.ROApr 7, 2026

Simulation-Driven Evolutionary Motion Parameterization for Contact-Rich Granular Scooping with a Soft Conical Robotic Hand

Authors: Yongliang Wang, Cristian C. Beltran-Hernandez, Tomoya Takahashi, Masashi Hamaya

Organizations: OMRON SINIC X Corporation, Tokyo, Japan. · Department of Artificial Intelligence, Bernoulli Institute, Faculty of Science and Engineering, University of Groningen, The Netherlands

Abstract

Tool-based scooping is vital in robot-assisted tasks, enabling interaction with objects of varying sizes, shapes, and material states. Recent studies have shown that flexible, reconfigurable soft robotic end-effectors can adapt their shape to maintain consistent contact with container surfaces during scooping, improving efficiency compared to rigid tools. These soft tools can adjust to varying container sizes and materials without requiring complex sensing or control. However, the inherent compliance and complex deformation behavior of soft robotics introduce significant control complexity that limits practical applications. To address this challenge, this paper presents the development of a physics-based simulation model of a deformable soft conical robotic hand that captures its passive reconfiguration dynamics and enables systematic trajectory optimization for scooping tasks. We propose a novel physics-based simulation approach that accurately models the soft tool's morphing behavior from flat sheets to adaptive conical structures, combined with an evolutionary strategy framework that automatically optimizes scooping trajectories without manual parameter tuning. We validate the optimized trajectories through both simulation and real-robot experiments. The results demonstrate strong generalization and successfully address a range of challenging tasks previously beyond the reach of existing approaches. Videos of our experiments are available online: https://sites.google.com/view/scoopsh

Figures & tables

Fig. 1: Contact-rich granular scooping system: The framework couples real-world execution (left) with its digital twin in simulation (right). From RGB-D perception and feature extraction, container geometry is abstracted into seed trajectories. In simulation, the covariance matrix adaptation evolution strategy (CMA-ES) optimizes both trajectory and hand roll angle, producing the best solution. This optimized strategy is then directly transferred to the real soft hand without additional tuning, as demonstrated in the 10-ball scooping task.

Fig. 2: The real soft-hand [ 13 ] rolling from $20^{\circ}$ to $120^{\circ}$ in $20^{\circ}$ increments.

Panel of Figure 3 — Fig. 3: The top row shows, from left to right, the Model View, Wireframe View, Mesh View, and Inertia View. The bottom row illustrates the hand rolling from $20^{\circ}$ to $120^{\circ}$ in $20^{\circ}$ increments, highlighting its soft and deformable properties.

Fig. 4: Overall framework of our system: from RGB–D perception and ring–compass abstraction, through seed trajectory generation and parameterization, to trajectory optimization with CMA-ES, enabling transfer of optimized scooping strategies from simulation to the real robot.

Fig. 5: From left to right: anchor skeleton, parabolic fit in uv-plane, and piecewise-parabola trajectory.

Fig. 6: Perception pipeline: RGB–D input to ring–compass representation for trajectory planning.

Fig. 7: Generation and refinement of scooping trajectories. From left to right: initial trajectory seeded from the ring–compass skeleton, intermediate optimization process, and final optimized trajectory.

Fig. 8: Containers used in experiments: $12$ objects in three sizes (large, medium, small), including regular and irregular shapes. (Width x Height)

Action Parameterization	Parameters	Scooped
Action Parameterization	Parameters	Big	Medium	Small
Initialization	$(k,a,b,\Delta angle)$	0	0	0
No $k_{1}$	$(k_{2},a_{1},b_{1},\Delta angle)$	2	1	1
No $k_{2}$	$(k_{1},a_{1},b_{1},\Delta angle)$	6	5	5
No $a_{1}$	$(k_{1},k_{2},b_{1},\Delta angle)$	6	7	8
No $b_{1}$	$(k_{1},k_{2},a_{1},\Delta angle)$	8	9	9

TABLE I: Scooping results under different action parameterizations across 3 container sizes in the 10 balls task.

Task	Trial 1	2	3	4	5	6	7	8	9	10	Average
10-balls	10	7	8	8	5	9	5	9	9	9	7.9
20-balls	16	16	16	13	15	15	14	14	14	16	14.9
50g-rice (g)	48.23	47.99	48.12	48.27	47.82	47.92	49.06	48.13	48.60	48.38	48.25

TABLE II: Scooping results over 10 trials for three tasks in a real robot.

Panel of Figure 11 — Fig. 9: Action sequence of the SH executing the optimized trajectory transferred from simulation to reality. The figure illustrates key phases of the scooping process, including rolling SH, insertion, and lifting. The left is the 20-ball task, and the right is the 50g-rice task.

Robotic Manipulation

Robotic manipulation studies how robots perceive, plan, and control physical interactions to grasp, move, assemble, and use objects.

See all 1,685 papers

Learning Adaptive Force Control for Contact-Rich Sample Scraping with Heterogeneous Materials
Mar 11, 2026Cenk Cetin, Shreyas Pouli, Gabriella PizzutoContact-Rich Robotic Manipulation Force-Controlled Robotic Manipulation
School of Computer Science & Informatics, University of Liverpool, UK. · Department of Chemistry, University of Liverpool, UK.
Functional Force-Aware Retargeting from Virtual Human Demos to Soft Robot Policies
Apr 1, 2026Uksang Yoo, Mengjia Zhu, Evan Pezent +8Motion Retargeting Robot Imitation Learning
The Robotics Institute, Carnegie Mellon University · Meta Reality Labs
Optimize, Learn, Refine: Whole-Body Grasping and Pick-and-Throw with a Spiral Soft Robot
Sep 22, 2026Marwah Basuhai, Tingcong Liu, Ibrahim Alsarraj +2Robotic Grasping Whole-Body Robotic Manipulation

Mar 11, 2026J/K move · Enter open · S save

Learning Adaptive Force Control for Contact-Rich Sample Scraping with Heterogeneous Materials

Cenk Cetin, Shreyas Pouli, Gabriella Pizzuto

School of Computer Science & Informatics, University of Liverpool, UK. · Department of Chemistry, University of Liverpool, UK.

Contact-Rich Robotic Manipulation Force-Controlled Robotic Manipulation Laboratory Automation Robotic RL

Open paper AskPDF arXiv

The increasing demand for accelerated scientific discovery, driven by global challenges, highlights the need for advanced AI-driven robotics. Deploying robotic chemists in human-centric labs is key for the next horizon of autonomous discovery, as complex tasks still demand the dexterity of human scientists. Robotic manipulation in this context is uniquely challenged by handling diverse chemicals (granular, powdery, or viscous liquids), under varying lab conditions. For example, humans use spatulas for scraping materials from vial walls. Automating this process is challenging because it goes beyond simple robotic insertion tasks and traditional lab automation, requiring the execution of fine-granular movements within a constrained environment (the sample vial). Our work proposes an adaptive control framework to address this, relying on a low-level Cartesian impedance controller for stable and compliant physical interaction and a high-level reinforcement learning agent that learns to dynamically adjust interaction forces at the end-effector. The agent is guided by perception feedback, which provides the material's location. We first created a task-representative simulation environment with a Franka Research 3 robot, a scraping tool, and a sample vial containing heterogeneous materials. To facilitate the learning of an adaptive policy and model diverse characteristics, the sample is modelled as a collection of spheres, where each sphere is assigned a unique dislodgement force threshold, which is procedurally generated using Perlin noise. We train an agent to autonomously learn and adapt the optimal contact wrench for a sample scraping task in simulation and then successfully transfer this policy to a real robotic setup. Our method was evaluated across five different material setups, outperforming a fixed-wrench baseline by an average of 10.9%.

Abstract

Robotic Manipulation

Explore similar work

Learning Adaptive Force Control for Contact-Rich Sample Scraping with Heterogeneous Materials

Functional Force-Aware Retargeting from Virtual Human Demos to Soft Robot Policies

Optimize, Learn, Refine: Whole-Body Grasping and Pick-and-Throw with a Spiral Soft Robot