The Effect of Haptic Guidance on Learning a Hybrid Rhythmic-Discrete Motor Task

Dissociating haptic feedback from physical assistance does not improve motor performance

In robots for motor rehabilitation and sports training, haptic assistance typically provides both mechanical guidance and task-relevant information. With the natural human tendency to minimise metabolic cost, mechanical guidance may however prevent efficient short term learning and retention. In this work, we explore the effect of providing haptic feedback to the not active hand during a tracking task. We test four types of haptic feedback: task- or error-related information, no information and irrelevant information. The results show that feedback provided to the hand not carrying out the tracking task did not improve task performance. However, irrelevant information to the task worsened performance, and negatively influenced the participants' perception of helpfulness, assistance, likability and predictability.

Interaction with a reactive partner improves learning in contrast to passive guidance

Many tasks such as physical rehabilitation, vehicle co-piloting or surgical training, rely on physical assistance from a partner. While this assistance may be provided by a robotic interface, how to implement the necessary haptic support to help improve performance without impeding learning is unclear. In this paper, we study the influence of haptic interaction on the performance and learning of a shared tracking task. We compare in a tracking task the interaction with a human partner, the trajectory guidance traditionally used in training robots, and a robot partner yielding human-like interaction. While trajectory guidance resulted in the best performance during training, it dramatically reduced error variability and hindered learning. In contrast, the reactive human and robot partners did not impede the adaptation and allowed the subjects to learn without modifying their movement patterns. Moreover, interaction with a human partner was the only condition that demonstrated an improvement in retention and transfer learning compared to a subject training alone. These results reveal distinctly different learning behaviour in training with a human compared to trajectory guidance, and similar learning between the robotic partner and human partner. Therefore, for movement assistance and learning, algorithms that react to the user's motion and change their behaviour accordingly are better suited.

Towards functional robotic training: motor learning of dynamic tasks is enhanced by haptic rendering but hampered by arm weight support

Background

Current robot-aided training allows for high-intensity training but might hamper the transfer of learned skills to real daily tasks. Many of these tasks, e.g., carrying a cup of coffee, require manipulating objects with complex dynamics. Thus, the absence of somatosensory information regarding the interaction with virtual objects during robot-aided training might be limiting the potential benefits of robotic training on motor (re)learning. We hypothesize that providing somatosensory information through the haptic rendering of virtual environments might enhance motor learning and skill transfer. Furthermore, the inclusion of haptic rendering might increase the task realism, enhancing participants’ agency and motivation. Providing arm weight support during training might also enhance learning by limiting participants’ fatigue.

Methods

We conducted a study with 40 healthy participants to evaluate how haptic rendering and arm weight support affect motor learning and skill transfer of a dynamic task. The task consisted of inverting a virtual pendulum whose dynamics were haptically rendered on an exoskeleton robot designed for upper limb neurorehabilitation. Participants trained with or without haptic rendering and with or without weight support. Participants’ task performance, movement strategy, effort, motivation, and agency were evaluated during baseline, short- and long-term retention. We also evaluated if the skills acquired during training transferred to a similar task with a shorter pendulum.

Results

We found that haptic rendering significantly increases participants’ movement variability during training and the ability to synchronize their movements with the pendulum, which is correlated with better performance. Weight support also enhances participants’ movement variability during training and reduces participants’ physical effort. Importantly, we found that training with haptic rendering enhances motor learning and skill transfer, while training with weight support hampers learning compared to training without weight support. We did not observe any significant differences between training modalities regarding agency and motivation during training and retention tests.

Conclusion

Haptic rendering is a promising tool to boost robot-aided motor learning and skill transfer to tasks with similar dynamics. However, further work is needed to find how to simultaneously provide robotic assistance and haptic rendering without hampering motor learning, especially in brain-injured patients.

Trial registrationhttps://clinicaltrials.gov/show/NCT04759976

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-022-00993-w.

Haptic Training: Which Types Facilitate (re)Learning of Which Motor Task and for Whom? Answers by a Review

The use of robots has attracted researchers to design numerous haptic training methods to support motor learning. However, investigations of new methods yielded inconclusive results regarding their effectiveness to enhance learning due to the diversity of tasks, haptic designs, participants' skill level, and study protocols. In this review, we developed a taxonomy to identify generalizable findings out of publications on haptic training. In the taxonomy, we grouped the results of studies on healthy learners based on participants' skill level and tasks' characteristics. Our inspection of included studies revealed that: i) Performance-enhancing haptic methods were beneficial for novices, ii) Training with haptics was as effective as training with other feedback modalities, and iii) Performance-enhancing and performance-degrading haptic methods were useful for the learning of temporal and spatial aspects, respectively. We also observed that these findings are in line with results from robot-aided neurorehabilitation studies on patients. Our review suggests that haptic training can be effective to foster learning, especially when the information cannot be provided with other feedback modalities. We believe the findings from the taxonomy constitute a general guide, which can assist researchers when designing studies to investigate the effectiveness of haptics on learning different tasks.

Promoting Motor Variability During Robotic Assistance Enhances Motor Learning of Dynamic Tasks

Despite recent advances in robot-assisted training, the benefits of haptic guidance on motor (re)learning are still limited. While haptic guidance may increase task performance during training, it may also decrease participants' effort and interfere with the perception of the environment dynamics, hindering somatosensory information crucial for motor learning. Importantly, haptic guidance limits motor variability, a factor considered essential for learning. We propose that Model Predictive Controllers (MPC) might be good alternatives to haptic guidance since they minimize the assisting forces and promote motor variability during training. We conducted a study with 40 healthy participants to investigate the effectiveness of MPCs on learning a dynamic task. The task consisted of swinging a virtual pendulum to hit incoming targets with the pendulum ball. The environment was haptically rendered using a Delta robot. We designed two MPCs: the first MPC-end-effector MPC-applied the optimal assisting forces on the end-effector. A second MPC-ball MPC-applied its forces on the virtual pendulum ball to further reduce the assisting forces. The participants' performance during training and learning at short- and long-term retention tests were compared to a control group who trained without assistance, and a group that trained with conventional haptic guidance. We hypothesized that the end-effector MPC would promote motor variability and minimize the assisting forces during training, and thus, promote learning. Moreover, we hypothesized that the ball MPC would enhance the performance and motivation during training but limit the motor variability and sense of agency (i.e., the feeling of having control over their movements), and therefore, limit learning. We found that the MPCs reduce the assisting forces compared to haptic guidance. Training with the end-effector MPC increases the movement variability and does not hinder the pendulum swing variability during training, ultimately enhancing the learning of the task dynamics compared to the other groups. Finally, we observed that increases in the sense of agency seemed to be associated with learning when training with the end-effector MPC. In conclusion, training with MPCs enhances motor learning of tasks with complex dynamics and are promising strategies to improve robotic training outcomes in neurological patients.

Haptic human-human interaction does not improve individual visuomotor adaptation

Haptic interaction between two humans, for example, a physiotherapist assisting a patient regaining the ability to grasp a cup, likely facilitates motor skill acquisition. Haptic human–human interaction has been shown to enhance individual performance improvement in a tracking task with a visuomotor rotation perturbation. These results are remarkable given that haptically assisting or guiding an individual rarely benefits their individual improvement when the assistance is removed. We, therefore, replicated a study that reported that haptic interaction between humans was beneficial for individual improvement for tracking a target in a visuomotor rotation perturbation. In addition, we tested the effect of more interaction time and a stronger haptic coupling between the partners on individual improvement in the same task. We found no benefits of haptic interaction on individual improvement compared to individuals who practised the task alone, independent of interaction time or interaction strength.

A Haptic Shared-Control Architecture for Guided Multi-Target Robotic Grasping

Although robotic telemanipulation has always been a key technology for the nuclear industry, little advancement has been seen over the last decades. Despite complex remote handling requirements, simple mechanically linked master-slave manipulators still dominate the field. Nonetheless, there is a pressing need for more effective robotic solutions able to significantly speed up the decommissioning of legacy radioactive waste. This paper describes a novel haptic shared-control approach for assisting a human operator in the sort and segregation of different objects in a cluttered and unknown environment. A three-dimensional scan of the scene is used to generate a set of potential grasp candidates on the objects at hand. These grasp candidates are then used to generate guiding haptic cues, which assist the operator in approaching and grasping the objects. The haptic feedback is designed to be smooth and continuous as the user switches from a grasp candidate to the next one, or from one object to another one, avoiding any discontinuity or abrupt changes. To validate our approach, we carried out two human-subject studies, enrolling 15 participants. We registered an average improvement of 20.8%, 20.1%, and 32.5% in terms of completion time, linear trajectory, and perceived effectiveness, respectively, between the proposed approach and standard teleoperation.

Caring About the Human Operator: Haptic Shared Control for Enhanced User Comfort in Robotic Telemanipulation

Haptic shared control enables a human operator and an autonomous controller to share the control of a robotic system using haptic active constraints. It has been used in robotic teleoperation for different purposes, such as navigating along paths minimizing the torques requested to the manipulator or avoiding possibly dangerous areas of the workspace. However, few works have focused on using these ideas to account for the user's comfort. In this article, we present an innovative haptic-enabled shared control approach aimed at minimizing the user's workload during a teleoperated manipulation task. Using an inverse kinematic model of the human arm and the rapid upper limb assessment (RULA) metric, the proposed approach estimates the current user's comfort online. From this measure and an a priori knowledge of the task, we then generate dynamic active constraints guiding the users towards a successful completion of the task, along directions that improve their posture and increase their comfort. Studies with human subjects show the effectiveness of the proposed approach, yielding a 30% perceived reduction of the workload with respect to using standard guided human-in-the-loop teleoperation.

Rowing Simulator Modulates Water Density to Foster Motor Learning

Although robot-assisted training is present in various fields such as sports engineering and rehabilitation, provision of training strategies that optimally support individual motor learning remains as a challenge. Literature has shown that guidance strategies are useful for beginners, while skilled trainees should benefit from challenging conditions. The Challenge Point Theory also supports this in a way that learning is dependent on the available information, which serves as a challenge to the learner. So, learning can be fostered when the optimal amount of information is given according to the trainee's skill. Even though the framework explains the importance of difficulty modulation, there are no practical guidelines for complex dynamic tasks on how to match the difficulty to the trainee's skill progress. Therefore, the goal of this study was to determine the impact on learning of a complex motor task by a modulated task difficulty scheme during the training sessions, without distorting the nature of task. In this 3-day protocol study, we compared two groups of naïve participants for learning a sweep rowing task in a highly sophisticated rowing simulator. During trainings, groups received concurrent visual feedback displaying the requested oar movement. Control group performed the task under constant difficulty in the training sessions. Experimental group's task difficulty was modulated by changing the virtual water density that generated different heaviness of the simulated water-oar interaction, which yielded practice variability. Learning was assessed in terms of spatial and velocity magnitude errors and the variability for these metrics. Results of final day tests revealed that both groups reduced their error and variability for the chosen metrics. Notably, in addition to the provision of a very well established visual feedback and knowledge of results, experimental group's variable training protocol with modulated difficulty showed a potential to be advantageous for the spatial consistency and velocity accuracy. The outcomes of training and test runs indicate that we could successfully alter the performance of the trainees by changing the density value of the virtual water. Therefore, a follow-up study is necessary to investigate how to match different density values to the skill and performance improvement of the participants.

Do we need complex rehabilitation robots for training complex tasks?

One key question in motor learning is how the complex tasks in daily life - those that require coordinated movements of multiple joints - should be trained. Often, complex tasks are directly taught as a whole, even though training of simple movement components before training the entire movement has been shown to be more effective for particularly complex tasks ("part-whole transfer paradigm"). The important implication of the part-whole transfer paradigm, e.g. on the field of rehabilitation robotics, is that training of most complex tasks could be simplified and, subsequently, devices used to train can become simpler and more affordable. In this way, robot-assisted rehabilitation could become more accessible. However, often the last step in the training process is forgotten: the recomposition of several simple movement components to a complete complex movement. Therefore, at least for the last training step, a complex rehabilitation device may be required.In a pilot study, we wanted to investigate if a complex robotic device (e.g. an exoskeleton robot with many degrees of freedom), such as the ARMin rehabilitation robot, is really beneficial for training the coordination between several simpler movement components or if training using visual feedback would lead to equal benefits. In a study, involving 16 healthy participants, who were instructed in a complex rugby motion, we could show first trends on the following two aspects: i) the part-whole transfer paradigm seems to hold true and therefore, simple robots might be used for training movement primitives. ii) Visual feedback does not seem to have the same potential, at least in healthy humans, to replace visuo-haptic guidance for movement recomposition of complex tasks. Therefore, complex rehabilitation robots seem to be beneficial for training complex real-life tasks.

Virtual Reality Environments and Haptic Strategies to Enhance Implicit Learning and Motivation in Robot-Assisted Training

Motivation plays a crucial role in motor learning and neurorehabilitation. Participants' motivation could decline to a point where they may stop training when facing a very difficult task. Conversely, participants may perform well and consider the training boring if the task is too easy. In this paper, we present a combination of a virtual reality environment with different robotic training strategies that modify task functional difficulty to enhance participants' motivation. We employed a pneumatically driven robotic stepper as a haptic interface. We first evaluated the use of disturbance observers as acceleration controllers to provide high robustness to varying system parameters, unmodeled dynamics and unknown disturbances associated with pneumatic control. The locomotor task to be learned in the virtual reality environment consisted of steering a recumbent bike to follow a desired path by changing the movement frequency of the dominant leg. The motor task was specially designed to engage implicit learning -i.e., learning without conscious recognition of what is learned. A haptic assistance strategy was developed in order to reduce the task functional difficulty during practice. In a feasibility study with eight healthy participants, we found that the haptic assistance provided by the robotic device successfully contributed to improve task performance during training, especially for less skilled participants. Furthermore, we found a negative correlation between participants' motivation and performance error when training with haptic assistance, suggesting that haptic assistance has a great potential to enhance motivation during motor training.

Haptic Error Modulation Outperforms Visual Error Amplification When Learning a Modified Gait Pattern

Robotic algorithms that augment movement errors have been proposed as promising training strategies to enhance motor learning and neurorehabilitation. However, most research effort has focused on rehabilitation of upper limbs, probably because large movement errors are especially dangerous during gait training, as they might result in stumbling and falling. Furthermore, systematic large movement errors might limit the participants' motivation during training. In this study, we investigated the effect of training with novel error modulating strategies, which guarantee a safe training environment, on motivation and learning of a modified asymmetric gait pattern. Thirty healthy young participants walked in the exoskeletal robotic system Lokomat while performing a foot target-tracking task, which required an increased hip and knee flexion in the dominant leg. Learning the asymmetric gait pattern with three different strategies was evaluated: (i) No disturbance: no robot disturbance/guidance was applied, (ii) haptic error amplification: unsafe and discouraging large errors were limited with haptic guidance, while haptic error amplification enhanced awareness of small errors relevant for learning, and (iii) visual error amplification: visually observed errors were amplified in a virtual reality environment. We also evaluated whether increasing the movement variability during training by adding randomly varying haptic disturbances on top of the other training strategies further enhances learning. We analyzed participants' motor performance and self-reported intrinsic motivation before, during and after training. We found that training with the novel haptic error amplification strategy did not hamper motor adaptation and enhanced transfer of the practiced asymmetric gait pattern to free walking. Training with visual error amplification, on the other hand, increased errors during training and hampered motor learning. Participants who trained with visual error amplification also reported a reduced perceived competence. Adding haptic disturbance increased the movement variability during training, but did not have a significant effect on motor adaptation, probably because training with haptic disturbance on top of visual and haptic error amplification decreased the participants' feelings of competence. The proposed novel haptic error modulating controller that amplifies small task-relevant errors while limiting large errors outperformed visual error augmentation and might provide a promising framework to improve robotic gait training outcomes in neurological patients.

The self-organization of ball bouncing

The hybrid rhythmic ball-bouncing task considered in this study requires a participant to hit a ball in a virtual environment by moving a paddle in the real environment. It allows for investigation of the online visual control of action in humans. Changes in gravity acceleration in the virtual environment affect the ball dynamics and modify the ball-paddle system limit cycle. These changes are shown to be accurately reproduced through simulation by a model integrating continuous information-movement couplings between the ball trajectory and the paddle trajectory, giving rise to a resonance-tuning phenomenon. On the contrary, the tested models integrating only intermittent sensorimotor couplings were unable to replicate the observed human behavior. Results suggest that the visual control of action is achieved online, in a prospective way. Human rhythmic motor control would benefit from the timing and phase control emerging from the low-level continuous coupling between the central pattern generator and the visual perception of the ball trajectory. This control strategy, which precludes the need for internal clock and explicit environmental representation, is also able to explain the empirical result that the bounces tend to converge toward a passive stability regime during human ball bouncing.

Neural circuits activated by error amplification and haptic guidance training techniques during performance of a timing-based motor task by healthy individuals

To promote motor learning, robotic devices have been used to improve subjects' performance by guiding desired movements (haptic guidance-HG) or by artificially increasing movement errors to foster a more rapid learning (error amplification-EA). To better understand the neurophysiological basis of motor learning, a few studies have evaluated brain regions activated during EA/HG, but none has compared both approaches. The goal of this study was to investigate using fMRI which brain networks were activated during a single training session of HG/EA in healthy adults learning to play a computerized pinball-like timing task. Subjects had to trigger a robotic device by flexing their wrist at the correct timing to activate a virtual flipper and hit a falling ball towards randomly positioned targets. During training with HG/EA, subjects' timing errors were decreased/increased, respectively, by the robotic device to delay or accelerate their wrist movement. The results showed that at the beginning of the training period with HG/EA, an error-detection network, including cerebellum and angular gyrus, was activated, consistent with subjects recognizing discrepancies between their intended actions and the actual movement timing. At the end of the training period, an error-detection network was still present for EA, while a memory consolidation/automatization network (caudate head and parahippocampal gyrus) was activated for HG. The results indicate that training movement with various kinds of robotic input relies on different brain networks. Better understanding the neurophysiological underpinnings of brain processes during HG/EA could prove useful for optimizing rehabilitative movement training for people with different patterns of brain damage.

Continuous Movement Tracking Performance for Predictable and Unpredictable Tasks with Vibrotactile Feedback

The purpose of this paper was to determine human movement tracking performance in response to vibrotactile feedback tracking for predictable and unpredictable continuous movement tasks. Thirteen subjects performed elbow flexion/extension and knee flexion/extension continuous movement tracking tasks while receiving tactile stimulation proportional to limb joint position error. Subjects followed 0.2-2.0 Hz desired movements for predictable tasks (single sinusoid) and unpredictable tasks (combination of three sinusoids). Tactile stimulation reaction times at the forearm to induce elbow flexion/extension and at the shank to induce knee flexion/extension were also recorded. Results of frequency tracking showed that 100 percent of participants correctly tracked unpredictable tasks at all frequencies, but only 60-80 percent of participants correctly tracked predictable tasks at frequencies less than 1 Hz and only 20-60 percent of participants correctly tracked predictable tasks at frequencies greater than 1 Hz. Subjects had less phase lag for predictable tasks than for unpredictable tasks. Reaction times at the forearm were 379 ms and at the shank 437 ms. These findings suggest that continuous vibrotactile feedback based on position errors may not be the most effective means of training higher frequency human movements and serve to inform future vibrotactile feedback design related to training human limb movements for predictable and unpredictable tasks.

Model of rhythmic ball bouncing using a visually controlled neural oscillator

The present paper investigates the sensory-driven modulations of central pattern generator dynamics that can be expected to reproduce human behavior during rhythmic hybrid tasks. We propose a theoretical model of human sensorimotor behavior able to account for the observed data from the ball-bouncing task. The novel control architecture is composed of a Matsuoka neural oscillator coupled with the environment through visual sensory feedback. The architecture's ability to reproduce human-like performance during the ball-bouncing task in the presence of perturbations is quantified by comparison of simulated and recorded trials. The results suggest that human visual control of the task is achieved online. The adaptive behavior is made possible by a parametric and state control of the limit cycle emerging from the interaction of the rhythmic pattern generator, the musculoskeletal system, and the environment.NEW & NOTEWORTHY The study demonstrates that a behavioral model based on a neural oscillator controlled by visual information is able to accurately reproduce human modulations in a motor action with respect to sensory information during the rhythmic ball-bouncing task. The model attractor dynamics emerging from the interaction between the neuromusculoskeletal system and the environment met task requirements, environmental constraints, and human behavioral choices without relying on movement planning and explicit internal models of the environment.

Implicit guidance to stable performance in a rhythmic perceptual-motor skill

Feedback about error or reward is regarded essential for aiding learners to acquire a perceptual-motor skill. Yet, when a task has redundancy and the mapping between execution and performance outcome is unknown, simple error feedback does not suffice in guiding the learner toward the optimal solutions. The present study developed and tested a new means of implicitly guiding learners to acquire a perceptual-motor skill, rhythmically bouncing a ball on a racket. Due to its rhythmic nature, this task affords dynamically stable solutions that are robust to small errors and noise, a strategy that is independent from actively correcting error. Based on the task model implemented in a virtual environment, a time-shift manipulation was designed to shift the range of ball-racket contacts that achieved dynamically stable solutions. In two experiments, subjects practiced with this manipulation that guided them to impact the ball with more negative racket accelerations, the indicator for the strategy with dynamic stability. Subjects who practiced under normal conditions took longer time to acquire this strategy, although error measures were identical between the control and experimental groups. Unlike in many other haptic guidance or adaptation studies, the experimental groups not only learned, but also maintained the stable solution after the manipulation was removed. These results are a first demonstration that more subtle ways to guide the learner to better performance are needed especially in tasks with redundancy, where error feedback may not be sufficient.