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

Cenk Cetin; Shreyas Pouli; Gabriella Pizzuto

cs.ROMar 11, 2026

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

Authors: Cenk Cetin, Shreyas Pouli, Gabriella Pizzuto

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

Abstract

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%.

Figures & tables

Fig. 1 : The proposed control architecture for autonomous sample scraping. A high-level RL policy receives a visual state ( $s_{\text{p}}=[v_{1},v_{2},v_{3}]^{T}$ ), with $v_{i}=[c_{ix},c_{iy},c_{iz},p_{i}]^{T}$ denoting each cluster’s centroid and residue percentage) and the robot’s Cartesian state with the external wrench ( $x,\dot{x},F_{\text{ext}}$ ). The policy outputs a hybrid action command ( $\boldsymbol{a}_{t}=[f_{x}^{c},\tau_{y}^{c},z^{D}]^{T}$ ) at 10 Hz, defining a desired force, torque, and position in the z-axis. This command is used as a goal to the Cartesian impedance controller running at 500 Hz, which generates the compliant joint torques ( $\tau_{c}$ ).

Fig. 2 : The sequential phases of the autonomous in-vial sample scraping task: ( left ) tool insertion with lateral force $F_{x}$ and torque $\tau_{y}$ applied against the vial wall, with RGB-D camera capturing visual state $s_{p}$ ; ( centre ) RL policy adapts $F_{x}$ , $\tau_{y}$ , and vertical position $z^{D}$ based on real-time perception feedback $s_{p}$ , controlling both contact wrench and scraping direction; ( right ) final upward sweep with adjusted $F_{x}$ to clear residual material, with dislodged particles settled at the bottom of the vial.

Fig. 3 : The proposed perception pipeline. From RGB-D input, YOLO [ 18 ] localises the vial, segmented via GrabCut [ 19 ] . Depth filtering isolates front-facing material and colour-based filtering removes the spatula. K-means clustering over a region of interest (ROI) then yields the material clusters, each represented by a centroid ( $c_{x},c_{y},c_{z}$ ) and residue percentage ( $p$ ).

Fig. 4 : A dynamic depth threshold method to detect the presence of material closest to the camera. The threshold is calculated by interpolating between the closest and furthest depth measurements based on a user-defined ratio ( e.g. , a ratio of 0.5 would correspond to the front half of the vial.)

Panel of Table 5 — TABLE I : Qualitative results for the five material samples, from the original sample to those obtained after scraping by the baseline (fixed wrench) and RL (adaptive wrench) methods.

Panel of Figure 6 — Fig. 5 : An overview of the simulation environment and real robot setup, including the robotic manipulator (Franka Research 3), the scraping tool and the vial.

Fig. 6 : An overview of the simulated heterogeneous materials. Left: Particles are randomly generated with varying hardness levels, represented by shades of green (lighter green indicates softer, easier-to-dislodge particles; darker green indicates harder particles requiring greater force). Right: After scraping, successfully dislodged particles turn red, while remaining green particles represent material yet to be removed.

Material	Robot Success (%)	Human Success (%)	Relative Success (%)
Liquid Dough	$56.8\pm 6.8$	$90.1\pm 4.5$	$63.0\pm 6.5$
Liquid Cornflour	$63.4\pm 9.9$	$92.5\pm 2.9$	$68.4\pm 9.2$
Dried Cornflour	$83.3\pm 8.9$	$94.1\pm 1.9$	$88.6\pm 9.6$
Crystalline Salt	$59.0\pm 13.6$	$96.2\pm 4.2$	$61.2\pm 14.3$
Crystalline Sugar	$7.5\pm 2.7$	$19.1\pm 8.5$	$41.0\pm 8.4$

TABLE II : Results for fixed wrench profile (baseline method). Success rates averaged over 5 trials.

Method	Acc.	Prec.	Rec.	Spec.	F1
Spatula	79.91%	78.94%	82.91%	76.75%	80.88%
No Spatula	77.19%	74.09%	86.71%	66.74%	79.91%
Filtered Spatula	64.71%	79.40%	40.04%	89.55%	53.24%

TABLE III : Results for the perception method, averaged over 20 trials for crystalline materials, were evaluated using accuracy, precision, recall, specificity, and F1-score. The methods are: (1) the vial with the spatula (Spatula), (2) the vial without the spatula (No Spatula), and (3) the vial with the spatula algorithmically removed from the image (Filtered Spatula).

Material	Robot Success (%)	Human Success (%)	Relative Success (%)
Liquid Dough	$66.4\pm 9.4$	$88.5\pm 5.5$	$74.9\pm 8.7$
Liquid Cornflour	$57.2\pm 11.1$	$82.1\pm 13.3$	$70.0\pm 10.3$
Dried Cornflour	$84.2\pm 2.6$	$89.8\pm 2.2$	$93.8\pm 1.8$
Crystalline Salt	$62.0\pm 18.1$	$85.7\pm 16.2$	$71.4\pm 10.3$
Crystalline Sugar	$19.9\pm 12.6$	$28.5\pm 15.0$	$66.4\pm 11.2$

TABLE IV : Results for our method (RL for adaptive wrench generation). Success rates averaged over 5 trials.

Contact-Rich Robotic Manipulation

Contact-rich robotic manipulation studies how robots plan and control physical interactions such as grasping, insertion, pushing, and dexterous object handling.

See all 413 papers

Dec 24, 2025

Tracing Energy Flow: Learning Tactile-based Grasping Force Control to Reduce Slippage in Dynamic Object Interaction

Regulating grasping force to reduce slippage during dynamic object interaction remains a fundamental challenge in robotic manipulation, especially when objects are manipulated by multiple rolling contacts, have unknown properties (such as mass or surface conditions), and when external sensing is unreliable. In contrast, humans can quickly regulate grasping force by touch, even without visual cues. Inspired by this ability, we aim to enable robotic hands to rapidly explore objects and learn tactile-driven grasping force control under motion and limited sensing. We propose a physics-informed energy abstraction that models the object as a virtual energy container. The inconsistency between the fingers' applied power and the object's retained energy provides a physically grounded signal for inferring slip-aware stability. Building on this abstraction, we employ model-based learning and planning to efficiently model energy dynamics from tactile sensing and perform real-time grasping force optimization. Experiments in both simulation and hardware demonstrate that our method can learn grasping force control from scratch within minutes, effectively reduce slippage, and extend grasp duration across diverse motion-object pairs, all without relying on external sensing or prior object knowledge. (Video: https://youtu.be/l3TJV29Mo6w)

Cheng-Yu Kuo, Hirofumi Shin, Takamitsu Matsubara

Graduated School of Science and Technology, Nara Institute of Science and Technology, Nara 630-0192, Japan · Honda R&D, Ltd., Saitama 351-0114, Japan

Robotic Grasping Contact-Rich Robotic Manipulation Force-Controlled Robotic Manipulation Feedback Control Tactile Sensing for Robotic Manipulation

Ask PDF

Oct 22, 2026

FORCA: A Force-Aware Coding Agent for Contact-Rich Robot Manipulation

Coding agents can translate language instructions into robot programs. For contact-rich manipulation, a key challenge is to generate and refine control logic that responds to physical interaction. We present FORCA, a coding agent that takes visual observations and wrist force/torque (F/T) measurements as input to write Python programs for contact probing and force-responsive motion. Between executions, the agent inspects force records alongside visual outcomes to diagnose failures and revise both motion and force-feedback control logic. FORCA requires no demonstrations, task-specific training, or simulation. Across five real-robot tasks, we demonstrate its effectiveness in three forms of contact-rich manipulation: improving manipulation accuracy, adapting to changes in object or surface properties, and regulating contact under surface variation and object motion. Compared with a vision-only variant of the same agent, FORCA improves grasp accuracy and task success and reduces contact loss in wiping and peeling.

Xiao Tan, Yixuan Wang, Zhexi Wang +1

All authors are with the University of Illinois Urbana-Champaign, Urbana, IL, USA. · Zhexi Wang is also with Tsinghua University, Beijing, China.

Contact-Rich Robotic Manipulation Code Generation for Robotics Robotic Manipulation Force-Controlled Robotic Manipulation

Ask PDF

Mar 16, 2026

Master Micro Residual Correction with Adaptive Tactile Fusion and Force-Mixed Control for Contact-Rich Manipulation

Robotic contact-rich and fine-grained manipulation remains a significant challenge due to complex interaction dynamics and the competing requirements of multi-timescale control. While current visual imitation learning methods excel at long-horizon planning, they often fail to perceive critical interaction cues like friction variations or incipient slip, and struggle to balance global task coherence with local reactive feedback. To address these challenges, we propose M2-ResiPolicy, a novel Master-Micro residual control architecture that synergizes high-level action guidance with low-level correction. The framework consists of a Master-Guidance Policy (MGP) operating at 10 Hz, which generates temporally consistent action chunks via a diffusion-based backbone and employs a tactile-intensity-driven adaptive fusion mechanism to dynamically modulate perceptual weights between vision and touch. Simultaneously, a high-frequency (60 Hz) Micro-Residual Corrector (MRC) utilizes a lightweight GRU to provide real-time action compensation based on TCP wrench feedback. This policy is further integrated with a force-mixed PBIC execution layer, effectively regulating contact forces to ensure interaction safety. Experiments across several demanding tasks including fragile object grasping and precision insertion, demonstrate that M2-ResiPolicy significantly outperforms standard Diffusion Policy (DP) and state-of-the-art Reactive Diffusion Policy (RDP), achieving a 93% damage-free success rate in chip grasping and superior force regulation stability.

Xingting Li, Yifan Xie, Han Liu +4

Shenzhen Ubiquitous Data Enabling Key Lab, Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China. · School of Electronics and Communication Engineering, Sun Yat-sen University, Shenzhen 518107, China. · School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639956, Singapore.

Robotic Manipulation Contact-Rich Robotic Manipulation Residual Policy Learning Diffusion Policy Force-Controlled Robotic Manipulation Tactile Sensing for Robotic Manipulation

Ask PDF

Abstract

Contact-Rich Robotic Manipulation

Explore similar work

Tracing Energy Flow: Learning Tactile-based Grasping Force Control to Reduce Slippage in Dynamic Object Interaction

FORCA: A Force-Aware Coding Agent for Contact-Rich Robot Manipulation

Master Micro Residual Correction with Adaptive Tactile Fusion and Force-Mixed Control for Contact-Rich Manipulation