Pith. sign in

REVIEW 4 major objections 5 minor 66 references

PulseRide: A Robotic Wheelchair for Personalized Exertion Control with Human-in-the-Loop Reinforcement Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A motorized manual wheelchair that watches heart rate and ECG learns when to help push, and for ten participants it kept users in a personalized moderate-activity zone up to 71.7 percent longer than manual propulsion.

desk verdict A plausible new integration of HR/ECG with RL assist that deserves a referee, but its headline numbers rest on an undescribed manual baseline and circular reward/evaluation thresholds. read the letter →

arxiv 2506.05056 v1 pith:SZF7EVOC submitted 2025-06-05 cs.RO cs.HC

Azizul Zahid , Bibek Poudel , Danny Scott , Jason Scott , Scott Crouter , Weizi Li , Sai Swaminathan This is my paper
classification cs.ROcs.HC
keywords assistiveroboticsreinforcementlearningwheelchairhuman-in-the-loopECGheartratezonesexertioncontrolpersonalizedassistance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

PulseRide is a manual wheelchair with two small motors that can push the wheels, plus a heart-rate and ECG monitor. The paper's central claim is that a reinforcement learning controller, trained for twenty minutes on each user, learns when and for how long to switch the motors on so that the user's heart rate stays inside a personally calibrated moderate-activity zone. In tests with ten participants on slate and carpet, the authors report that individual users remained in that zone up to 71.7 percent longer than with an unassisted manual wheelchair, that muscle contractions dropped by an average of 41.86 percent, and that fatigue was delayed. The aim is to offer a middle path between power wheelchairs, which discourage activity, and manual wheelchairs, which strain the shoulders and arms.

What carries the argument

The load-bearing mechanism is a three-part control loop. The first part is a frozen convolutional ECG encoder, trained on four people to classify heart signals into low, medium, or high activity and then stripped of its classification head so it outputs eight features per three-second interval. The second part is a Deep Q-Network, a deep network that estimates the future reward value of each action, whose observation stacks two seconds of heart rate, velocity, previous motor action, and those eight ECG features; its only actions are motor on or motor off. The third part is a hand-shaped reward function that uses each user's personalized heart-rate and velocity thresholds, gathered in a 135-second pre-training session, to reward time spent in the moderate zone and penalize extreme heart rate, excessive velocity, or applying assistance at already-high speed. Together these parts let the system adapt assistance to an individual without explicit environment programming.

What would settle it

Train the ECG feature extractor on four people, then run the full PulseRide protocol on a fresh group of participants and measure how well the frozen feature extractor classifies their heart signals into low, medium, and high activity; if held-out accuracy is near chance, or if the per-user time in the moderate zone is no better than manual wheeling in a replication, the central claim fails.

Watch

Extended reading notes

Core claim

The discovery this paper seeks to establish is that a live physiological signal can act as the control input for adaptive wheelchair assistance. During a short pre-training phase, the system measures each user's own heart-rate and velocity ranges at self-chosen low, medium, and high effort, producing personalized thresholds. A Deep Q-Network then receives, every two seconds, heart rate, wheelchair velocity, the previous motor action, and an eight-dimensional encoding of the last three seconds of ECG produced by a convolutional network that was first trained to classify physical activity level. The network chooses only whether to turn the motors on or off, and the reward function rewards staying inside the moderate zone while penalizing over-exertion and excessive speed. The reported result is that this policy kept users' heart rates in the moderate zone longer than manual wheeling, with an individual best of 71.7 percent more time, while reducing muscle contractions and shifting electromyographic (EMG) mean frequency in the direction of less fatigue; the policy also generalized from training on slate to testing on carpet.

Load-bearing premise

The entire assistance decision depends on eight features that a machine-learning model extracts from the heart signal, and that model was trained on only four people who were not part of the ten-person study; if those features do not remain meaningful for a new user, the learned policy is acting on uninformative inputs.

Editorial extensions

If this is right

  • If the central claim holds, wheelchair users can accumulate moderate cardiovascular exercise during ordinary trips instead of choosing between full manual effort and a fully powered chair.
  • Because the policy trained only on slate still kept heart rates in the moderate zone on carpet, a single training session may transfer to other indoor surfaces with different rolling resistance.
  • Personalized thresholds derived from each user's own propulsion effort mean the same hardware can serve users with very different fitness levels without manual tuning.
  • Lower push frequency and reduced muscle contractions address the repetitive-strain injuries that make manual wheelchair use unsustainable for many people.
  • The retrofit attaches to a stock manual wheelchair and costs under $500, so the approach is potentially accessible outside specialized clinical hardware.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • These are editorial extensions beyond the paper's own claims.
  • Because PulseRide can only turn the motors on or off and has no braking, it cannot actively lower heart rate on downhill or coasting stretches; real-world gains may be smaller wherever gravity does the work.
  • Both the encoder training pool, four people, and the main study, ten people, were people without disabilities, so the ECG features and the learned policy still need to be revalidated in wheelchair users, especially those with altered cardiovascular responses.
  • A natural next experiment would keep the same reward structure but allow continuous motor thrust or several assist levels, since the binary action likely under-uses the physiological information; the 71.7 percent figure should also be reported as a distribution, since it is the best individual user's gain rather than a group average.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces PulseRide, an add-on motorized wheelchair system that uses human-in-the-loop deep Q-learning to provide individualized push assistance based on heart rate, ECG, and velocity. The authors formalize the control problem as an MDP, train a CNN encoder on ECG data from four separate participants, and evaluate the system with ten non-disabled participants on slate and carpet surfaces. The headline results are that PulseRide keeps users in their personalized moderate heart-rate zone up to 71.7% longer than a manual wheelchair, reduces muscle contractions by 41.86% on slate and 24.94% on carpet, and appears to delay fatigue based on EMG mean-frequency trends. The paper also claims that PulseRide is the first wheelchair system to offer personalized, adaptive assistance balancing physical effort with mobility needs.

Significance. If the causal claims were established, PulseRide would be a meaningful contribution: a low-cost (~$500) retrofitted wheelchair with real-time physiological feedback and personalized assistance, addressing a real gap between manual and powered wheelchairs. The MDP formulation, the two-phase personalization, and the use of a frozen ECG encoder are sensible system components, and the authors are transparent about the pilot nature of the study and the decision to test with participants without disabilities. However, the current evidence is preliminary: the manual-wheelchair comparison is not described in the methods, no inferential statistics are reported, and the fatigue evidence rests on a single participant's slope plus an undefined aggregate measure. The contribution is therefore an interesting system description and pilot evaluation, rather than a validated effectiveness claim.

major comments (4)
  1. [Section 4.2] The Procedure section describes only a PulseRide sequence (introduction, rest, pre-training, 20-minute training, 5-minute test on slate and carpet) and never specifies a manual-wheelchair condition. Yet Section 5 repeatedly compares PulseRide against 'manual wheelchair' (Figures 5-11), and the abstract quantifies gains of 71.7% and 41.86%. Without a description of the manual condition—its order relative to PulseRide training, randomization, instructions, duration, and whether the same personalized zones were used—these differences cannot be attributed to the assistive policy rather than to fatigue, practice, or instruction effects. This omission is load-bearing for the central claim.
  2. [Section 5.1-5.3, Figures 8-10] No inferential statistics are reported for any comparison. Figure 10 shows only average percentages with standard-deviation error bars; Figures 8 and 9 show per-user counts without paired tests or effect sizes. With n=10, individual outliers (e.g., Users 3, 7, and 8 in Figures 5-6) can dominate averages. The paper should report paired statistical tests or explicit effect sizes with confidence intervals for time-in-zone, contraction counts, and MNF slopes, or alternatively downgrade the causal claims to descriptive pilot observations.
  3. [Section 5.4, Figure 11] The fatigue-delay claim is insufficiently supported. Figure 11B shows MNF slopes for one participant only; Figure 11C, described as 'MNF variability across all participants,' does not present per-participant slopes or a group-level slope analysis. The text states that PulseRide 'consistently demonstrated more positive MNF variability' but does not define how this variability is computed or whether it corresponds to a fatigue-relevant slope. Without group-level MNF slope statistics, the conclusion that PulseRide delays fatigue is not established.
  4. [Algorithm 1 and Section 4.3.2] The primary outcome, proportion of time in the moderate heart-rate zone, is computed with the same personalized λ thresholds that define the reward function in Algorithm 1. The evaluation metric is therefore aligned with the training objective by construction. This does not invalidate the system goal, but the manuscript should acknowledge this circularity and validate the zone boundaries against an independent standard (e.g., heart-rate reserve, ACSM moderate-intensity ranges, or participant-rated RPE) before claiming that PulseRide maintains 'moderate activity' as a health-relevant construct.
minor comments (5)
  1. [Algorithm 1] Algorithm 1 contains the condition 'velocity = vel_cv' but vel_cv is never defined in the text or tables; this appears to be a typographical or omitted definition.
  2. [Section 3.3] The ECG encoder is trained on only 4 participants (303 samples), yet it is used as a core observation for the 10 study participants. Since the policy also uses heart rate and velocity, the central zone-maintenance result may be robust, but the generalization of the encoder to new users should be demonstrated or explicitly acknowledged as a limitation.
  3. [Section 4.2.1] The pre-training description says participants push at 'moderate and transition' levels, while Section 3.1 says three perceived exertion levels (low, medium, and high); please reconcile these descriptions.
  4. [Figure 11C] The caption states that error bars represent standard error of the mean, but no sample size or definition of 'MNF variability' is given; please specify the statistic and the number of participants.
  5. [Section 5.2] The phrase 'distinct clusters ... below 50.85%' is not the result of a cluster analysis; please rephrase to avoid implying a formal clustering method.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the headline outcomes are empirical measurements, not algebraic consequences of fitted thresholds.

full rationale

The claimed derivation chain is an empirical HITL-RL experiment, not a closed-form prediction. The personalized lambda thresholds from pre-training do enter both Algorithm 1's reward and the 'moderate zone' evaluation, but the reported quantities (71.7% zone time, 41.86% contraction reduction, MNF slope) are observed outcomes of the trained policy and manual trials; the paper's own Figures 8-11 show substantial inter-user variability and outlier users (User 3, User 10), so the outcomes are not forced by the reward definition. The ECG encoder is an auxiliary representation learned on separate participants and is a transfer assumption, not a circular reuse. The only self-citation ([47], 'Similar reward functions are commonly used in DC motor control problems') is peripheral and not load-bearing. The absence of a described manual-wheelchair condition in Section 4.2 and the lack of inferential statistics are serious methodological and correctness risks, but they are confounds, not circular derivations. Therefore no significant circularity is established.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The system introduces no new physical entities. It relies on per-user fitted thresholds, hand-tuned reward weights, and unverified transferability of a small-data ECG encoder. These assumptions, rather than any new physical ingredient, carry the central claims.

free parameters (5)
  • User-specific heart rate thresholds (lambda1h, lambda2h, lambda3h) = Per user, not reported numerically
    Derived in pre-training from 45-second pushes at perceived low, medium, and high exertion; define the moderate, transition, and extreme zones used in both the reward and the evaluation.
  • User-specific velocity thresholds (lambda1v, lambda2v, lambda3v) = Per user, not reported
    Derived alongside heart rate thresholds in pre-training; used to penalize high or reverse velocity in the reward function.
  • Reward function coefficients (0.5, 0.5, 3, 2, 1.2, 1.4, 0.6) = Hand-chosen constants
    Weights in Algorithm 1 set the shape of the reward; no sensitivity analysis or ablation is provided.
  • DQN and encoder hyperparameters = Table 1 values
    Batch sizes, learning rate, epsilon schedule, discount factor, and target network update ratio are chosen by the authors with no tuning procedure or ablations.
  • vel_cv = Undefined
    Appears in Algorithm 1 as a failure condition (velocity = vel_cv) but is never defined in the text or tables.
assumptions (4)
  • domain assumption The 135-second pre-training yields stable, meaningful personalized thresholds for each user.
    Section 4.2.1: noisy or unstable thresholds would affect both the reward the policy optimizes and the metric used to evaluate success.
  • ad hoc to paper The ECG encoder trained on 4 separate participants generalizes to new users.
    Section 3.3: the encoder is frozen and used as a feature extractor for the RL policy, but transferability is not evaluated on the study participants.
  • domain assumption Able-bodied participants are a valid proxy for wheelchair users in this system evaluation.
    Section 4.1 and Section 6.3: the authors acknowledge this is a proof of concept, but it remains load-bearing for any generalization to the target population.
  • domain assumption DQN with 600 training interactions learns a useful policy.
    Section 6.2: the authors note data scarcity and rely on encoder features and 2-second timesteps; no convergence guarantees for this regime are established.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PulseRide: A Robotic Wheelchair for Personalized Exertion Control with Human-in-the-Loop Reinforcement Learning." pith.science (2026). https://pith.science/paper/SZF7EVOC

@misc{pith2026250605056,
  author       = {Pith},
  title        = {Pith review of: PulseRide: A Robotic Wheelchair for Personalized Exertion Control with Human-in-the-Loop Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SZF7EVOC}},
  note         = {Machine review of arXiv:2506.05056}
}
read the original abstract

Maintaining an active lifestyle is vital for quality of life, yet challenging for wheelchair users. For instance, powered wheelchairs face increasing risks of obesity and deconditioning due to inactivity. Conversely, manual wheelchair users, who propel the wheelchair by pushing the wheelchair's handrims, often face upper extremity injuries from repetitive motions. These challenges underscore the need for a mobility system that promotes activity while minimizing injury risk. Maintaining optimal exertion during wheelchair use enhances health benefits and engagement, yet the variations in individual physiological responses complicate exertion optimization. To address this, we introduce PulseRide, a novel wheelchair system that provides personalized assistance based on each user's physiological responses, helping them maintain their physical exertion goals. Unlike conventional assistive systems focused on obstacle avoidance and navigation, PulseRide integrates real-time physiological data-such as heart rate and ECG-with wheelchair speed to deliver adaptive assistance. Using a human-in-the-loop reinforcement learning approach with Deep Q-Network algorithm (DQN), the system adjusts push assistance to keep users within a moderate activity range without under- or over-exertion. We conducted preliminary tests with 10 users on various terrains, including carpet and slate, to assess PulseRide's effectiveness. Our findings show that, for individual users, PulseRide maintains heart rates within the moderate activity zone as much as 71.7 percent longer than manual wheelchairs. Among all users, we observed an average reduction in muscle contractions of 41.86 percent, delaying fatigue onset and enhancing overall comfort and engagement. These results indicate that PulseRide offers a healthier, adaptive mobility solution, bridging the gap between passive and physically taxing mobility options.

Figures

Figures reproduced from arXiv: 2506.05056 by the authors.

Figure 1
Figure 1. Comparison of wheelchair assistance paradigms. Manual wheelchair (top) provides no assistance and requires [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (A) PulseRide’s reinforcement learning framework, using inputs such as encoded ECG, heart rate, velocity, and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (A) Activity classifier consisting of a Convolutional Neural Network (CNN) encoder and a Multi-Layered Perceptron [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: (A) Placement of electromyography (EMG) sensor [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Heart rate and velocity distributions across users [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Heart rate and velocity distributions across users [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 9
Figure 9. Figure 9: A total number of user muscle contractions per [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 8
Figure 8. Figure 8: Total number of muscle contractions for each user [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 10
Figure 10. Figure 10: Time spent in heart rate zones during tests (er [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: (A) Raw electromyography (EMG) signal with a [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

66 extracted references · 57 canonical work pages

  1. [1]

    Acekare. 2023. Yomper+ and Enjo: Power add-on for manual wheelchairs. https: //www.acekare.com/power-add-on-manual-wheelchair-yomper-enjo/ Accessed: 2024-11-02

    work page 2023

  2. [2]

    Seyler, Thomas Bruckner, Judith Munzinger, Thomas Grieser, Hans J

    Michael Akbar, Gabriel Balean, Manuela Brunner, Thorsten M. Seyler, Thomas Bruckner, Judith Munzinger, Thomas Grieser, Hans J. Gerner, and Markus Loew

    work page

  3. [3]

    Alber USA. 2023. e-motion: Active drive wheels for manual wheelchairs. https: //www.alber-usa.com/us/products/active-drives/e-motion/ Accessed: 2024-11- 02

    work page 2023

  4. [4]

    Mudasir Saleem Andrabi, Mercy Mumba, Betty Key, and Robert Motl. 2022. Physical activity programs for cardiovascular outcomes in community wheelchair users: A systematic review. Frontiers in Rehabilitation Sciences 3 (2022), 1007778

    work page 2022

  5. [5]

    Mahsa Asheghan, Mohammad Taghi Hollisaz, Taher Taheri, Hadi Kazemi, and Amidoddin Khatibi Aghda. 2016. The prevalence of carpal tunnel syndrome among long-term manual wheelchair users with spinal cord injury: A cross- sectional study. J. Spinal Cord Med. 39, 3 (May 2016), 265. https://doi.org/10. 1179/2045772315Y.0000000033

    work page 2016

  6. [6]

    Boninger, Alicia M

    Michael L. Boninger, Alicia M. Koontz, Sue Ann Sisto, Trevor A. Dyson-Hudson, Michael Chang, Robert Price, and Rory A. Cooper. 2005. Pushrim biomechanics and injury prevention in spinal cord injury: recommendations based on CULP- SCI investigations. J. Rehabil. Res. Dev. 42, 3 (May 2005), Suppl. https://doi.org/ 10.1682/jrrd.2004.08.0103 arXiv:16195959

    work page doi:10.1682/jrrd.2004.08.0103 2005

  7. [7]

    Frank W Booth, Scott E Gordon, Christian J Carlson, and Marc T Hamilton. 2000. Waging war on modern chronic diseases: primary prevention through exercise biology. Journal of applied physiology 88, 2 (2000), 774–787

    work page 2000

  8. [8]

    Johannes BJ Bussmann, Ulrich W Ebner-Priemer, and Jochen Fahrenberg. 2009. Ambulatory activity monitoring: Progress in measurement of activity, posture, and specific motion patterns in daily life. European Psychologist 14, 2 (2009), 142–152

    work page 2009

Show all 66 references
  1. [9]

    Sotirios Chatzidimitriadis and Konstantinos Sirlantzis. 2022. Deep Reinforcement Learning for Autonomous Navigation in Robotic Wheelchairs. In International Conference on Pattern Recognition and Artificial Intelligence . Springer, 271–282

    2022

  2. [10]

    R. A. Cooper, L. A. Quatrano, P. W. Axelson, W. Harlan, M. Stineman, B. Franklin, J. S. Krause, J. Bach, H. Chambers, E. Y. Chao, M. Alexander, and P. Painter. 1999. Research on physical activity and health among people with disabilities: a con- sensus statement. J. Rehabil. R...

    1999

  3. [11]

    Elaine H Coulter, Philippa M Dall, Lynn Rochester, Jon P Hasler, and Malcolm H Granat. 2011. Development and validation of a physical activity monitor for use on a wheelchair. Spinal cord 49, 3 (2011), 445–450

    2011

  4. [12]

    Rachel E Cowan, Stephanie L Silveira, Tina Helle, Uffe Læssøe, Kirstine Rosenbeck Gøeg, Jette Bangshaab, and Robert W Motl. 2022. Lifestyle physical activity in manual wheelchair users–an overlooked public health opportunity. Spinal Cord 60, 2 (2022), 190–192

    2022

  5. [13]

    Dalyan, D

    M. Dalyan, D. D. Cardenas, and B. Gerard. 1999. Upper extremity pain after spinal cord injury. Spinal Cord 37, 3 (March 1999), 191–195. https://doi.org/10.1038/sj. sc.3100802 arXiv:10213328

    1999 doi

  6. [14]

    Andrea L Dunn, Ross E Andersen, and John M Jakicic. 1998. Lifestyle physical ac- tivity interventions: History, short-and long-term effects, and recommendations. American journal of preventive medicine 15, 4 (1998), 398–412

    1998

  7. [15]

    Kara Edwards and Annie McCluskey. 2010. A survey of adult power wheelchair and scooter users. Disability and Rehabilitation: Assistive Technology 5, 6 (2010), 411–419

    2010

  8. [16]

    Arturo Escobedo, Anne Spalanzani, and Christian Laugier. 2013. Multimodal control of a robotic wheelchair: Using contextual information for usability im- provement. In 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 4262–4267

    2013

  9. [17]

    Guoxi Feng, Lucian Buşoniu, Thierry-Marie Guerra, and Sami Mohammad. 2019. Data-efficient reinforcement learning for energy optimization of power-assisted wheelchairs. IEEE Transactions on Industrial Electronics 66, 12 (2019), 9734–9744

    2019

  10. [18]

    Patrick Fougeyrollas and Luc Noreau. 2000. Long-term consequences of spinal cord injury on social participation: the occurrence of handicap situations. Dis- ability and rehabilitation 22, 4 (2000), 170–180

    2000

  11. [19]

    Fullerton, Jeffrey J

    Heather D. Fullerton, Jeffrey J. Borckardt, and Alan P. Alfano. 2003. Shoulder pain: a comparison of wheelchair athletes and nonathletic wheelchair users.Med. Sci. Sports Exerc. 35, 12 (Dec. 2003), 1958–1961. https://doi.org/10.1249/01.MSS. 0000099082.54522.55 arXiv:14652488

    2003

  12. [20]

    Shawn C Gendle, Mark Richardson, James Leeper, L Brent Hardin, J Matt Green, and Phillip A Bishop. 2012. Wheelchair-mounted accelerometers for measurement of physical activity. Disability and Rehabilitation: Assistive Technology 7, 2 (2012), 139–148

    2012

  13. [21]

    Giesbrecht, Jacqueline D

    Edward M. Giesbrecht, Jacqueline D. Ripat, Arthur O. Quanbury, and Juli- ette E. Cooper. 2009. Participation in community-based activities of daily living: comparison of a pushrim-activated, power-assisted wheelchair and a CHASE’25, Jun 24 2025 - Jun 26 2025, Manhattan, New Yo...

    2009 doi

  14. [22]

    Aditya Goil, Matthew Derry, and Brenna D. Argall. 2013. Using machine learning to blend human and robot controls for assisted wheelchair navigation. In 2013 IEEE 13th International Conference on Rehabilitation Robotics (ICORR) . 1–6. https: //doi.org/10.1109/ICORR.2013.6650454

    2013

  15. [23]

    JA Haisma, LHV Van Der Woude, HJ Stam, MP Bergen, TAR Sluis, and JBJ Bussmann. 2006. Physical capacity in wheelchair-dependent persons with a spinal cord injury: a critical review of the literature. Spinal cord 44, 11 (2006), 642–652

    2006

  16. [24]

    William L Haskell, I-Min Lee, Russell R Pate, Kenneth E Powell, Steven N Blair, Barry A Franklin, Caroline A Macera, Gregory W Heath, Paul D Thompson, and Adrian Bauman. 2007. Physical activity and public health: updated recommenda- tion for adults from the American College of...

    2007

  17. [25]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition . 770–778

    2016

  18. [26]

    GREGORY NV HEATH and Peter H Fentem. 1997. 8 physical activity among persons with disabilities—A public health perspective. Exercise and sport sciences reviews 25, 1 (1997), 195–234

    1997

  19. [27]

    Holloway, Andrew Symonds, Tatsuto Suzuki, Angela Gall, Peter Smitham, and Stephen Taylor

    Catherine S. Holloway, Andrew Symonds, Tatsuto Suzuki, Angela Gall, Peter Smitham, and Stephen Taylor. 2015. Linking wheelchair kinetics to glenohumeral joint demand during everyday accessibility activities. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. 2015:2478-81., ; (2015), A...

    2015 doi

  20. [28]

    Sungjin Hwang, Jiwoong Heo, Youngwug Cho, Jucheol Moon, Yushin Lee, Han- sung Kim, Jaehyuk Cha, and Kwanguk Kim. 2024. Transportation Mode Detection Technology to Predict Wheelchair Users’ Life Satisfaction in Seoul, South Korea. Proc. ACM Interact. Mob. Wearable Ubiquitous Te...

    2024 doi

  21. [29]

    TW Janssen, CA Van Oers, Edith P Rozendaal, EM Willemsen, A Peter Hollander, and LH Van der Woude. 1996. Changes in physical strain and physical capacity in men with spinal cord injuries. Medicine and science in sports and exercise 28, 5 (1996), 551–559

    1996

  22. [30]

    Robert T Kell, Gordon Bell, and Art Quinney. 2001. Musculoskeletal fitness, health outcomes and quality of life. Sports Medicine 31 (2001), 863–873

    2001

  23. [31]

    Diederik P Kingma. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  24. [32]

    Vijay Konda and John Tsitsiklis. 1999. Actor-critic algorithms.Advances in neural information processing systems 12 (1999)

    1999

  25. [33]

    Peng Liao, Walter Dempsey, Hillol Sarker, Syed Monowar Hossain, Mustafa Al’Absi, Predrag Klasnja, and Susan Murphy. 2018. Just-in-time but not too much: Determining treatment timing in mobile health. Proceedings of the ACM on interactive, mobile, wearable and ubiquitous techno...

    2018

  26. [34]

    Peng Liao, Kristjan Greenewald, Predrag Klasnja, and Susan Murphy. 2020. Per- sonalized HeartSteps: A Reinforcement Learning Algorithm for Optimizing Phys- ical Activity. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 4, 1, Article 18 (mar 2020), 22 pages. https://doi.o...

    2020 doi

  27. [35]

    Chen-Lung Lu, Zi-Yan Liu, Jui-Te Huang, Ching-I Huang, Bo-Hui Wang, Yi Chen, Nien-Hsin Wu, Hsueh-Cheng Wang, Laura Giarré, and Pei-Yi Kuo. 2021. Assistive navigation using deep reinforcement learning guiding robot with UWB/voice beacons and semantic feedbacks for blind and vis...

    2021

  28. [36]

    Biao Luo, Zhengke Wu, Fei Zhou, and Bing-Chuan Wang. 2024. Human-in-the- Loop Reinforcement Learning in Continuous-Action Space. IEEE Transactions on Neural Networks and Learning Systems 35, 11 (2024), 15735–15744. https: //doi.org/10.1109/TNNLS.2023.3289315

    2024

  29. [37]

    Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. 2015. Human-level control through deep reinforcement learning. nature 518, 7540 (2015), 529–533

    2015

  30. [38]

    Satoshi Muraki, Noriaki Tsunawake, Yasuaki Tahara, Sugaru Hiramatsu, and Masahiro Yamasaki. 2000. Multivariate analysis of factors influencing physical work capacity in wheelchair-dependent paraplegics with spinal cord injury. European Journal of Applied Physiology 81 (2000), 28–32

    2000

  31. [39]

    Mark S Nash, Rachel E Cowan, and Jochen Kressler. 2012. Evidence-based and heuristic approaches for customization of care in cardiometabolic syndrome after spinal cord injury. The journal of spinal cord medicine 35, 5 (2012), 278–292

    2012

  32. [40]

    Tom E Nightingale, Peter C Rouse, Dylan Thompson, and James LJ Bilzon. 2017. Measurement of physical activity and energy expenditure in wheelchair users: methods, considerations and future directions. Sports medicine-open 3 (2017), 1–16

    2017

  33. [41]

    Office of Disease Prevention, Health Promotion, et al . 2000. US Department of Health and Human Services: Healthy People 2010. http://www/health/gov/healthypeople/ (2000)

    2000

  34. [42]

    Satoshi Ohashi, Akira Shionoya, Keiu Harada, Masahito Nagamori, and Hisashi Uchiyama. 2022. Posture Estimation Using Surface Electromyography during Wheelchair Hand-Rim Operations. Sensors 22, 9 (2022). https://doi.org/10.3390/ s22093296

    2022

  35. [43]

    World Health Organization and United Nations Children’s Fund. 2022. Global report on assistive technology . World Health Organization

    2022

  36. [44]

    Russell R Pate, Michael Pratt, Steven N Blair, William L Haskell, Caroline A Macera, Claude Bouchard, David Buchner, Walter Ettinger, Gregory W Heath, Abby C King, et al. 1995. Physical activity and public health: a recommendation from the Centers for Disease Control and Preve...

    1995

  37. [45]

    Katrina L Piercy, Richard P Troiano, Rachel M Ballard, Susan A Carlson, Janet E Fulton, Deborah A Galuska, Stephanie M George, and Richard D Olson. 2018. The physical activity guidelines for Americans. Jama 320, 19 (2018), 2020–2028

    2018

  38. [46]

    Polomarco. 2023. ECG Classification: CNN-LSTM with Attention Mecha- nism. https://www.kaggle.com/code/polomarco/ecg-classification-cnn-lstm- attention-mechanism

    2023

  39. [47]

    Bibek Poudel, Thomas Watson, and Weizi Li. 2022. Learning to control dc motor for micromobility in real time with reinforcement learning. In 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC) . IEEE, 1248– 1254

    2022

  40. [48]

    L Qi, M Ferguson-Pell, Z Salimi, R Haennel, and A Ramadi. 2015. Wheelchair users’ perceived exertion during typical mobility activities. Spinal cord 53, 9 (2015), 687–691

    2015

  41. [49]

    W Jack Rejeski, Lawrence R Brawley, and Sally A Shumaker. 1996. Physical activity and health-related quality of life. Exercise and sport sciences reviews 24, 1 (1996), 71–108

    1996

  42. [50]

    D Rimaud, P Calmels, and X Devillard. 2005. Training programs in spinal cord injury. In Annales de readaptation et de medecine physique: revue scientifique de la Societe francaise de reeducation fonctionnelle de readaptation et de medecine physique, Vol. 48. 259–269

    2005

  43. [51]

    Sandra B Rodrigues, Luís Palermo de Faria, António M Monteiro, José Luís Lima, Tiago M Barbosa, and José A Duarte. 2022. EMG signal processing for the study of localized muscle fatigue—pilot study to explore the applicability of a novel method. International Journal of Environ...

    2022

  44. [52]

    Hye-Yeon Ryu, Je-Seong Kwon, Jeong-Hak Lim, A-Hyeon Kim, Su-Jin Baek, and Jong-Wook Kim. 2021. Development of an autonomous driving smart wheelchair for the physically weak. Applied Sciences 12, 1 (2021), 377

    2021

  45. [53]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  46. [54]

    Mandi Sehgal, Jeremy Jacobs, and Wendy S Biggs. 2021. Mobility assistive device use in older adults. American family physician 103, 12 (2021), 737–744

    2021

  47. [55]

    Richard S Sutton. 2018. Reinforcement learning: An introduction. A Bradford Book (2018)

    2018

  48. [56]

    Richard S Sutton, David McAllester, Satinder Singh, and Yishay Mansour. 1999. Policy gradient methods for reinforcement learning with function approximation. Advances in neural information processing systems 12 (1999)

    1999

  49. [57]

    Michelle L Tolerico, Dan Ding, Rory A Cooper, Donald M Spaeth, Shirley G Fitzgerald, Rosemarie Cooper, Annmarie Kelleher, and Michael L Boninger. 2007. Assessing mobility characteristics and activity levels of manual wheelchair users. Journal of Rehabilitation Research & Devel...

    2007

  50. [58]

    Mohd Razali Md Tomari, Yoshinori Kobayashi, and Yoshinori Kuno. 2012. Devel- opment of smart wheelchair system for a user with severe motor impairment. Procedia Engineering 41 (2012), 538–546

    2012

  51. [59]

    Anne Vuillemin, Stéphanie Boini, Sandrine Bertrais, Sabrina Tessier, Jean-Michel Oppert, Serge Hercberg, Francis Guillemin, and Serge Briançon. 2005. Leisure time physical activity and health-related quality of life. Preventive medicine 41, 2 (2005), 562–569

    2005

  52. [60]

    Alireza Fayazi, Hamed Saeidi, and Ardalan Vahidi

    Nianfeng Wan, S. Alireza Fayazi, Hamed Saeidi, and Ardalan Vahidi. 2014. Op- timal power management of an electric bicycle based on terrain preview and considering human fatigue dynamics. Proceedings of the American Control Con- ference, 3462–3467. https://doi.org/10.1109/ACC....

    2014

  53. [61]

    Darren ER Warburton, Norman Gledhill, and Arthur Quinney. 2001. The effects of changes in musculoskeletal fitness on health. Canadian journal of applied physiology 26, 2 (2001), 161–216

    2001

  54. [62]

    Darren ER Warburton, Crystal Whitney Nicol, and Shannon SD Bredin. 2006. Health benefits of physical activity: the evidence. Cmaj 174, 6 (2006), 801–809

    2006

  55. [63]

    Jacob O Wobbrock, Shaun K Kane, Krzysztof Z Gajos, Susumu Harada, and Jon Froehlich. 2011. Ability-based design: Concept, principles and examples. ACM Transactions on Accessible Computing (TACCESS) 3, 3 (2011), 1–27

    2011

  56. [64]

    World Health Organization. 2023. Assistive Technology. https://www.who.int/ en/news-room/fact-sheets/detail/assistive-technology Accessed: 2024-09-27. Received 6 January 2025; revised 31 March 2025; accepted 30 June 20259

    2023

  57. [2010]

    Prevalence of rotator cuff tear in paraplegic patients compared with controls. J. Bone Joint Surg. Am. 92, 1 (Jan. 2010), 23–30. https://doi.org/10.2106/JBJS.H. 01373 arXiv:20048092

    2010 doi

  58. [2017]

    arXiv preprint arXiv:1707.06347 (2017)

    Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 (2017)

    2017 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.