Sensor-Enhanced Smart Gripper Development for Automated Meat Processing.

Grasping and object manipulation have been considered key domains of Cyber-Physical Systems (CPS) since the beginning of automation, as they are the most common interactions between systems, or a system and its environment. As the demand for automation is spreading to increasingly complex fields of industry, smart tools with sensors and internal decision-making become necessities. CPS, such as robots and smart autonomous machinery, have been introduced in the meat industry in recent decades; however, the natural diversity of animals, potential anatomical disorders and soft, slippery animal tissues require the use of a wide range of sensors, software and intelligent tools. This paper presents the development of a smart robotic gripper for deployment in the meat industry. A comprehensive review of the available robotic grippers employed in the sector is presented along with the relevant recent research projects. Based on the identified needs, a new mechatronic design and early development process of the smart gripper is described. The integrated force sensing method based on strain measurement and magnetic encoders is described, including the adjacent laboratory and on-site tests. Furthermore, a combined slip detection system is presented, which relies on an optical flow-based image processing algorithm using the video feed of a built-in endoscopic camera. Basic user tests and application assessments are presented.

Force Control of a Haptic Flexible-Link Antenna Based on a Lumped-Mass Model.

Haptic organs are common in nature and help animals to navigate environments where vision is not possible. Insects often use slender, lightweight, and flexible links as sensing antennae. These antennae have a muscle-endowed base that changes their orientation and an organ that senses the applied force and moment, enabling active sensing. Sensing antennae detect obstacles through contact during motion and even recognize objects. They can also push obstacles. In all these tasks, force control of the antenna is crucial. The objective of our research is to develop a haptic robotic system based on a sensing antenna, consisting of a very lightweight and slender flexible rod. In this context, the work presented here focuses on the force control of this device. To achieve this, (a) we develop a dynamic model of the antenna that moves under gravity and maintains point contact with an object, based on lumped-mass discretization of the rod; (b) we prove the robust stability property of the closed-loop system using the Routh stability criterion; and (c) based on this property, we design a robust force control system that performs efficiently regardless of the contact point with the object. We built a mechanical device replicating this sensing organ. It is a flexible link connected at one end to a 3D force-torque sensor, which is attached to a mechanical structure with two DC motors, providing azimuthal and elevation movements to the antenna. Our experiments in contact situations demonstrate the effectiveness of our control method.

Neural Filtering of Physiological Tremor Oscillations to Spinal Motor Neurons Mediates Short-Term Acquisition of a Skill Learning Task.

The acquisition of a motor skill involves adaptations of spinal and supraspinal pathways to alpha motoneurons. In this study, we estimated the shared synaptic contributions of these pathways to understand the neural mechanisms underlying the short-term acquisition of a new force-matching task. High-density surface electromyography (HDsEMG) was acquired from the first dorsal interosseous (FDI; 7 males and 6 females) and tibialis anterior (TA; 7 males and 4 females) during 15 trials of an isometric force-matching task. For two selected trials (pre- and post-skill acquisition), we decomposed the HDsEMG into motor unit spike trains, tracked motor units between trials, and calculated the mean discharge rate and the coefficient of variation of interspike interval (COVISI). We also quantified the post/pre ratio of motor units' coherence within delta, alpha, and beta bands. Force-matching improvements were accompanied by increased mean discharge rate and decreased COVISI for both muscles. Moreover, the area under the curve within alpha band decreased by ∼22% (TA) and ∼13% (FDI), with no delta or beta bands changes. These reductions correlated significantly with increased coupling between force/neural drive and target oscillations. These results suggest that short-term force-matching skill acquisition is mediated by attenuation of physiological tremor oscillations in the shared synaptic inputs. Supported by simulations, a plausible mechanism for alpha band reductions may involve spinal interneuron phase-cancelling descending oscillations. Therefore, during skill learning, the central nervous system acts as a matched filter, adjusting synaptic weights of shared inputs to suppress neural components unrelated to the specific task.

Muscle contractile properties directly influence shared synaptic inputs to spinal motor neurons.

Alpha band oscillations in shared synaptic inputs to the alpha motor neuron pool can be considered an involuntary source of noise that hinders precise voluntary force production. This study investigated the impact of changing muscle length on the shared synaptic oscillations to spinal motor neurons, particularly in the physiological tremor band. Fourteen healthy individuals performed low-level dorsiflexion contractions at ankle joint angles of 90° and 130°, while high-density surface electromyography (HDsEMG) was recorded from the tibialis anterior (TA). We decomposed the HDsEMG into motor units spike trains and calculated the motor units' coherence within the delta (1-5 Hz), alpha (5-15 Hz), and beta (15-35 Hz) bands. Additionally, force steadiness and force spectral power within the tremor band were quantified. Results showed no significant differences in force steadiness between 90° and 130°. In contrast, alpha band oscillations in both synaptic inputs and force output decreased as the length of the TA was moved from shorter (90°) to longer (130°), with no changes in delta and beta bands. In a second set of experiments (10 participants), evoked twitches were recorded with the ankle joint at 90° and 130°, revealing longer twitch durations in the longer TA muscle length condition compared to the shorter. These experimental results, supported by a simple computational simulation, suggest that increasing muscle length enhances the muscle's low-pass filtering properties, influencing the oscillations generated by the Ia afferent feedback loop. Therefore, this study provides valuable insights into the interplay between muscle biomechanics and neural oscillations. KEY POINTS: We investigated whether changes in muscle length, achieved by changing joint position, could influence common synaptic oscillations to spinal motor neurons, particularly in the tremor band (5-15 Hz). Our results demonstrate that changing muscle length from shorter to longer induces reductions in the magnitude of alpha band oscillations in common synaptic inputs. Importantly, these reductions were reflected in the oscillations of muscle force output within the alpha band. Longer twitch durations were observed in the longer muscle length condition compared to the shorter, suggesting that increasing muscle length enhances the muscle's low-pass filtering properties. Changes in the peripheral contractile properties of motor units due to changes in muscle length significantly influence the transmission of shared synaptic inputs into muscle force output. These findings prove the interplay between muscle mechanics and neural adaptations.

Beyond the manual touch: situational-aware force control for increased safety in robot-assisted skullbase surgery.

PURPOSE: Skullbase surgery demands exceptional precision when removing bone in the lateral skull base. Robotic assistance can alleviate the effect of human sensory-motor limitations. However, the stiffness and inertia of the robot can significantly impact the surgeon's perception and control of the tool-to-tissue interaction forces. METHODS: We present a situational-aware, force control technique aimed at regulating interaction forces during robot-assisted skullbase drilling. The contextual interaction information derived from the digital twin environment is used to enhance sensory perception and suppress undesired high forces. RESULTS: To validate our approach, we conducted initial feasibility experiments involving a medical and two engineering students. The experiment focused on further drilling around critical structures following cortical mastoidectomy. The experiment results demonstrate that robotic assistance coupled with our proposed control scheme effectively limited undesired interaction forces when compared to robotic assistance without the proposed force control. CONCLUSIONS: The proposed force control techniques show promise in significantly reducing undesired interaction forces during robot-assisted skullbase surgery. These findings contribute to the ongoing efforts to enhance surgical precision and safety in complex procedures involving the lateral skull base.

Safe contact-based robot active search using Bayesian optimization and control barrier functions.

In robotics, active exploration and learning in uncertain environments must take into account safety, as the robot may otherwise damage itself or its surroundings. This paper presents a method for safe active search using Bayesian optimization and control barrier functions. As robot paths undertaken during sampling are continuous, we consider an informative continuous expected improvement acquisition function. To safely bound the contact forces between the robot and its surroundings, we leverage exponential control barrier functions, utilizing the derivative of the force in the contact model to increase robustness to uncertainty in the contact boundary. Our approach is demonstrated on a fully autonomous robot for ultrasound scanning of rheumatoid arthritis (RA). Here, active search is a critical component of ensuring high image quality. Furthermore, bounded contact forces between the ultrasound probe and the patient ensure patient safety and better scan quality. To the best of our knowledge, our results are both the first demonstration of safe active search on a fully autonomous robot for ultrasound scanning of rheumatoid arthritis and the first experimental evaluation of bounding contact forces in the context of medical robotics using control barrier functions. The results show that when search time is limited to less than 60 s, informative continuous expected improvement leads to a 92% success, a 13% improvement compared to expected improvement. Meanwhile, exponential control barrier functions can limit the force applied by the robot to under 5 N, even in cases where the contact boundary is specified incorrectly by -1 or +4 mm.

Repeated context-specific actions disrupt feedforward adjustments in motor commands in younger and older adults.

The flexibility of the motor system to adjust a planned action before or during the execution of the movement in response to sensory information is critical for preventing errors in motor control. As individuals age, this function declines, leading to an increased incidence of motor errors. Although sensory processing and cognitive decline are known contributors to this impairment, here, we test the hypothesis that repetition of context-specific planned actions interferes with the adjustment of feedforward motor commands. Younger and older participants were instructed to grasp and lift a T-shaped object with a concealed, off-sided center of mass and minimize its roll through anticipatory force control, relying predominantly on predictive model-driven planning (i.e., sensorimotor memories) developed through repeated lifts. We selectively manipulate the number of trial repeats with the center of mass on one side before switching it to the other side of the T-shaped object. The results showed that increasing the number of repetitions improved performance in manipulating an object with a given center of mass but led to increased errors when the object's center of mass was switched. This deleterious effect of repetition on feedforward motor adjustment was observed in younger and older adults. Critically, we show these effects on an internal model-driven motor planning task that relies predominantly on sensorimotor memory, with no differences in sensory inputs from the repetition manipulation. The findings indicate that feedforward motor adjustments are hampered by repetitive stereotyped planning and execution of motor behavior.NEW & NOTEWORTHY Adjusting planned actions in response to sensory stimuli degrades with age contributing to increased incidence of errors ranging from clumsy spills to catastrophic falls. Multiple factors likely contribute to age-related motor inflexibility, including sensory- and cognition-supporting system declines. Here, we present compelling evidence for repetition to disrupt feedforward adjusting of motor commands in younger and older adults, which suggests increases in stereotypy as a deleterious potentiator of motor control errors.

Experiment Comparative Analysis of Feed Rate with Velocity Control in Cutting Mono Crystalline Silicon Using a Diamond Wire Saw.

Fixed-diamond abrasive wire saw cutting is one of the most common methods for cutting hard and brittle materials. This process has unique advantages including a narrow kerf and the ability to use a relatively small cutting force. In the cutting process, controlling the main process parameters can improve the processing efficiency, obtaining a better processing surface roughness. This work designs the PI controller (Proportional-Integral controller) based on the reciprocating wire saw cutting process. The control objects are the workpiece feed rate and wire saw velocity, and the control objective is the normal cutting force. For the control trials, several reference values of various normal cutting forces were chosen. The effects of feed rate and saw velocity on the cutting surface finish and cutting time were investigated in this work using wire saw cutting analysis on a square monocrystalline silicon specimen. The results of this study showed that under a constant applied force of 2.5 N, the optimal feed rate of the diamond wire through the specimen could reduce cutting time by 42% while achieving a 60% improvement in the measured surface finish. Likewise, optimal control of the wire saw velocity could reduce cycle time by 18% with a 45% improvement in the surface finish. Consequently, the feed speed control is more effective than the wire saw velocity.

Postural Change of the Annual Cicada (Tibicen linnei) Helps Facilitate Backward Flight.

Cicadas are heavy fliers well known for their life cycles and sound production; however, their flight capabilities have not been extensively investigated. Here, we show for the first time that cicadas appropriate backward flight for additional maneuverability. We studied this flight mode using computational fluid dynamics (CFD) simulations based on three-dimensional reconstructions of high-speed videos captured in a laboratory. Backward flight was characterized by steep body angles, high angles of attack, and high wing upstroke velocities. Wing motion occurred in an inclined stroke plane that was fixed relative to the body. Likewise, the directions of the half-stroke-averaged aerodynamic forces relative to the body (local frame) were constrained in a narrow range (<20°). Despite the drastic difference of approximately 90° in body posture between backward and forward flight in the global frame, the aerodynamic forces in both flight scenarios were maintained in a similar direction relative to the body. The forces relative to the body were also oriented in a similar direction when observed during climbs and turns, although the body orientation and motions were different. Hence, the steep posture appropriated during backward flight was primarily utilized for reorienting both the stroke plane and aerodynamic force in the global frame. A consequence of this reorientation was the reversal of aerodynamic functions of the half strokes in backward flight when compared to forward flight. The downstroke generated propulsive forces, while the upstroke generated vertical forces. For weight support, the upstroke, which typically generates lesser forces in forward flight, is aerodynamically active in backward flight. A leading-edge vortex (LEV) was observed on the forewings during both half strokes. The LEV's effect, together with the high upstroke velocity, increased the upstroke's force contribution from 10% of the net forces in forward flight to 50% in backward flight. The findings presented in this study have relevance to the design of micro-aerial vehicles (MAVs), as backward flight is an important characteristic for MAV maneuverability or for taking off from vertical surfaces.

Effects of taxane-induced peripheral neuropathy on hand dexterity impairment: evaluation of quantitative and subjective assessments.

PURPOSE: Chemotherapy-induced peripheral neuropathy (CIPN) commonly involves hand dexterity impairment. However, the factors affecting hand dexterity impairment are unknown and there is currently no established treatment. The purpose of the current study was to clarify factors influencing hand dexterity impairment in taxane-induced peripheral neuropathy using subjective and objective assessments. METHODS: We assessed patient characteristics, treatment-related factors, subjective symptoms of CIPN (Patient Neurotoxicity Questionnaire [PNQ]), psychological symptoms, and upper limb dysfunction (Quick Disabilities of the Arm, Shoulder and Hand [Quick DASH]). Quantitative assessments were pinch strength, sensory threshold, hand dexterity impairment, and grip force control. Multiple regression analysis was performed using hand dexterity impairment as the dependent variable and age and PNQ, Quick DASH, and control of grip force as independent variables. RESULTS: Forty-three breast cancer patients were included in the analysis. Hand dexterity impairment in taxane-induced peripheral neuropathy patients was significantly correlated with age, grip force control, and PNQ sensory scores (p < 0.008). Multiple regression analysis demonstrated that PNQ sensory scores and grip force control were significantly associated with hand dexterity impairment (p < 0.01). CONCLUSION: Subjective symptoms (numbness and pain) and grip force control contributed to impaired hand dexterity in taxane-induced peripheral neuropathy.

Resistance band training with functional electrical stimulation improves force control capabilities in older adults: a preliminary study.

Resistance band training (RBT) with functional electrical stimulation (FES) may be an effective exercise regimen for improving age-related motor impairments. This preliminary study investigated the potential effects of bimanual RBT with FES on upper limb motor functions in older adults. This study randomly assigned 22 elderly people to the bimanual RBT with FES (Bi-RBT+FES) group and the RBT without FES (Bi-RBT) group. All participants performed isometric hand-grip force control tasks in unimanual (dominant and non-dominant) and bimanual conditions before and after four weeks of exercise for each group. We quantified the mean force, force accuracy, force variability, and force regularity at two targeted force levels (i.e., 10 % and 40 % of maximum voluntary contraction; MVC) to estimate changes in force control capabilities. The results revealed that the Bi-RBT+FES group demonstrated a greater force accuracy in the dominant hand at 10 % of MVC after training. Non-dominant hands in the Bi-RBT+FES group increased force accuracy at 40 % of MVC and reduced force variability collapsed across two targeted force levels. Both groups showed a decrease in force regularity after training. These preliminary results indicate that Bi-RBT+FES may be a viable option to facilitate functional recovery of the upper limbs in older adults.

Relative contribution of sensory and motor deficits on grip force control in patients with chronic stroke.

OBJECTIVE: This study aimed to characterize grasping behavior in static (weight-dependent modulation and stability of control) and dynamic (predictive control) aspects specifically focusing on the relative contribution of sensory and motor deficits to grip force control in patients with chronic stroke. METHODS: Twenty-four chronic stroke patients performed three manipulative tasks: five trials of 5-s grasp-lift-holding, 30-s static holding, and vertical dynamic/cyclic oscillation of holding the object. RESULTS: Exerted static grip force on the paretic side exhibited statistically greater than that on the non-paretic side. Spearman's rank correlation coefficient revealed that the contribution to static grip force control was larger in sensory deficits than in motor deficits. In addition, the sensory deficit is related to the reduced coupling between grip force and load force, suggesting difficulty in predictive control due to the absence of sensory feedback. CONCLUSIONS: Given that grip force control involves predictive feedforward and online feedback control, the evaluation of grip force might be an important and feasible evaluation manner for the assessment of sensorimotor control in patients post-stroke. SIGNIFICANCE: Detailed evaluation of grip force control would help to understand the mechanisms underlying hand dysfunction in stroke patients.

Role of arm reaching movement kinematics in friction perception at initial contact with smooth surfaces.

When manipulating objects, humans begin adjusting their grip force to friction within 100 ms of contact. During motor adaptation, subjects become aware of the slipperiness of touched surfaces. Previously, we have demonstrated that humans cannot perceive frictional differences when surfaces are brought in contact with an immobilised finger, but can do so when there is submillimeter lateral displacement or subjects actively make the contact movement. Similarly, in, we investigated how humans perceive friction in the absence of intentional exploratory sliding or rubbing movements, to mimic object manipulation interactions. We used a two-alternative forced-choice paradigm in which subjects had to reach and touch one surface followed by another, and then indicate which felt more slippery. Subjects correctly identified the more slippery surface in 87 ± 8% of cases (mean ± SD; n = 12). Biomechanical analysis of finger pad skin displacement patterns revealed the presence of tiny (<1 mm) localised slips, known to be sufficient to perceive frictional differences. We tested whether these skin movements arise as a result of natural hand reaching kinematics. The task was repeated with the introduction of a hand support, eliminating the hand reaching movement and minimising fingertip movement deviations from a straight path. As a result, our subjects' performance significantly declined (66 ± 12% correct, mean ± SD; n = 12), suggesting that unrestricted reaching movement kinematics and factors such as physiological tremor, play a crucial role in enhancing or enabling friction perception upon initial contact. KEY POINTS: More slippery objects require a stronger grip to prevent them from slipping out of hands. Grip force adjustments to friction driven by tactile sensory signals are largely automatic and do not necessitate cognitive involvement; nevertheless, some associated awareness of grip surface slipperiness under such sensory conditions is present and helps to select a safe and appropriate movement plan. When gripping an object, tactile receptors provide frictional information without intentional rubbing or sliding fingers over the surface. However, we have discovered that submillimeter range lateral displacement might be required to enhance or enable friction sensing. The present study provides evidence that such small lateral movements causing localised partial slips arise and are an inherent part of natural reaching movement kinematics.

A novel dual-arm adaptive cooperative control framework for carrying variable loads and active anti-overturning.

This paper aims to tackle the issue of carrying variable loads and disturbances in an impedance-based dual-arm robot. When robots are engaged in transportation tasks, deviations in trajectory due to changing loads and the risk of objects falling under external disturbances or unstable gripping can lead to mission failure. To address these issues, a novel Dual-Arm Adaptive Cooperative Control Framework (ACCF) is proposed. The ACCF is designed to ensure adherence to trajectory constraints in the presence of load variations and to actively respond to sliding or overturning caused by disturbances. The framework is user-friendly and robust, featuring a two-loop arrangement. The inner-loop incorporates an adaptive force control law to robustly control internal forces for dual-arm gripping. The outer-loop utilizes centralized impedance control, incorporating a fast gravity estimation scheme to compensate for trajectory deviations and an active anti-overturning scheme to resist sliding or overturning of objects during disturbances. Experimental evaluations and comparisons are conducted. The results successfully demonstrate the ACCF's adaptability under variable loads and external disturbances, providing a solution for practical dual-arm applications, such as transportation, in future scenarios.

Biomechanical Simulation of Orthodontic En-Bloc Retraction Comparing Compound Technique and Sliding Mechanics Using a HOSEA Robotic Device.

En-bloc retraction is a common procedure in orthodontic therapy. The application of palatal root torque moments is required to control incisor inclination during retraction, yet studies comparing forces and moments with respect to different mechanics are lacking. This study aimed to investigate the forces and moments during orthodontic en-bloc retraction using a robotic biomechanical simulation system, comparing two distinct approaches: (I) compound technique [stainless steel (SS) combined with nickel-titanium (NiTi)] using industrially pretorqued retraction-torque-archwires (RTA) in combination with NiTi closed coil springs; (II) conventional sliding mechanics using SS archwires with manually applied anterior twist bends in combination with elastic chains. Two dimensions (0.017" × 0.025" and 0.018" × 0.025") and ten archwires per group were investigated using 0.022" slot self-ligating brackets. Kruskal-Wallis tests with a significance level of α = 0.05 were conducted. The biomechanical simulation showed that en-bloc retraction was characterized by a series of tipping and uprighting movements, differing significantly regarding the examined mechanics. Collateral forces and moments occurred in all groups. Notably, RTA exhibited fewer extrusive forces. The most bodily movement was achieved with the compound technique and the 0.018" × 0.025" RTA. Sliding mechanics exhibited maximum palatal root torque moments of more than 20 Nmm, exceeding recommended values.

Human-robot planar co-manipulation of extended objects: data-driven models and control from human-human dyads.

Human teams are able to easily perform collaborative manipulation tasks. However, simultaneously manipulating a large extended object for a robot and human is a difficult task due to the inherent ambiguity in the desired motion. Our approach in this paper is to leverage data from human-human dyad experiments to determine motion intent for a physical human-robot co-manipulation task. We do this by showing that the human-human dyad data exhibits distinct torque triggers for a lateral movement. As an alternative intent estimation method, we also develop a deep neural network based on motion data from human-human trials to predict future trajectories based on past object motion. We then show how force and motion data can be used to determine robot control in a human-robot dyad. Finally, we compare human-human dyad performance to the performance of two controllers that we developed for human-robot co-manipulation. We evaluate these controllers in three-degree-of-freedom planar motion where determining if the task involves rotation or translation is ambiguous.

Effects of visual feedback and force level on bilateral ankle-dorsiflexion force control.

This study investigated the potential effects of visual feedback and force level on bilateral force control capabilities in the lower limbs. Thirty-nine healthy young adults performed bilateral ankle-dorsiflexion isometric force control tasks for different visual feedback conditions, including continuous visual feedback (CVF) and withdrawal of visual feedback (WVF), indicating the removal of visual feedback on force outputs during the task and force level conditions (i.e., 10 % and 40 % of the maximum voluntary contraction). Bilateral force control capabilities were estimated using force accuracy, variability, regularity, and absolute power in 0-4 Hz and interlimb coordination by cross-correlation with time lag and uncontrolled manifold (UCM) variables. Correlation analyses determined the relationship between changes in bilateral force control capabilities and interlimb coordination from the CVF to WVF conditions. The findings revealed better bilateral force control capabilities in the CVF condition as indicated by less force error, variability, regularity, absolute power in 0-4 Hz, and advanced interlimb force coordination. From CVF to WVF conditions, increased bad variability correlated with greater force control deficits. These findings suggest that visuomotor processing is an important resource for successful fine motor control in the lower limbs.

Human movement modifications induced by different levels of transparency of an active upper limb exoskeleton.

Active upper limb exoskeletons are a potentially powerful tool for neuromotor rehabilitation. This potential depends on several basic control modes, one of them being transparency. In this control mode, the exoskeleton must follow the human movement without altering it, which theoretically implies null interaction efforts. Reaching high, albeit imperfect, levels of transparency requires both an adequate control method and an in-depth evaluation of the impacts of the exoskeleton on human movement. The present paper introduces such an evaluation for three different "transparent" controllers either based on an identification of the dynamics of the exoskeleton, or on force feedback control or on their combination. Therefore, these controllers are likely to induce clearly different levels of transparency by design. The conducted investigations could allow to better understand how humans adapt to transparent controllers, which are necessarily imperfect. A group of fourteen participants were subjected to these three controllers while performing reaching movements in a parasagittal plane. The subsequent analyses were conducted in terms of interaction efforts, kinematics, electromyographic signals and ergonomic feedback questionnaires. Results showed that, when subjected to less performing transparent controllers, participants strategies tended to induce relatively high interaction efforts, with higher muscle activity, which resulted in a small sensitivity of kinematic metrics. In other words, very different residual interaction efforts do not necessarily induce very different movement kinematics. Such a behavior could be explained by a natural human tendency to expend effort to preserve their preferred kinematics, which should be taken into account in future transparent controllers evaluation.

A study on robot force control based on the GMM/GMR algorithm fusing different compensation strategies.

To address traditional impedance control methods' difficulty with obtaining stable forces during robot-skin contact, a force control based on the Gaussian mixture model/Gaussian mixture regression (GMM/GMR) algorithm fusing different compensation strategies is proposed. The contact relationship between a robot end effector and human skin is established through an impedance control model. To allow the robot to adapt to flexible skin environments, reinforcement learning algorithms and a strategy based on the skin mechanics model compensate for the impedance control strategy. Two different environment dynamics models for reinforcement learning that can be trained offline are proposed to quickly obtain reinforcement learning strategies. Three different compensation strategies are fused based on the GMM/GMR algorithm, exploiting the online calculation of physical models and offline strategies of reinforcement learning, which can improve the robustness and versatility of the algorithm when adapting to different skin environments. The experimental results show that the contact force obtained by the robot force control based on the GMM/GMR algorithm fusing different compensation strategies is relatively stable. It has better versatility than impedance control, and the force error is within ~±0.2 N.