Pith. sign in

REVIEW 3 major objections 6 minor 46 references

SpikingSoft: A Spiking Neuron Controller for Bio-inspired Locomotion with Soft Snake Robots

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read One double-threshold spiking neuron per segment, tuned by RL, can drive soft-snake locomotion by exciting the body's own oscillations, improving target-reaching success by 21.6 points and cutting time by 29%.

desk verdict A clever event-triggered spiking controller for a soft snake that shows real gains over vanilla RL, but the CPG baseline is a strawman and the smoothness claim is unquantified. read the letter →

arxiv 2501.19072 v2 pith:BHEHMK55 submitted 2025-01-31 cs.RO cs.LG

Chuhan Zhang , Cong Wang , Wei Pan , Cosimo Della Santina This is my paper · ORCID
classification cs.ROcs.LG
keywords spikingneuralnetworksdoublethresholdneuronsoftsnakerobotbio-inspiredlocomotionreinforcementlearningcentralpatterngeneratorevent-triggeredcontrolCosseratrodmodel
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

This paper tries to establish that a single spiking neuron per segment, with only two adjustable thresholds, can act as the entire low-level controller for a soft snake robot. The neuron takes the segment's deformation as input and emits torque spikes only when the deformation crosses a threshold, and the claim is that these event-triggered impulses lock onto and amplify the soft body's natural oscillations, producing forward slithering and turning without any preprogrammed gait curve. If true, the reinforcement-learning problem shrinks from outputting continuous torques to choosing two numbers per segment, making locomotion learning simpler and faster. The demonstrated evidence is simulated: in a target-reaching task, the SpikingSoft controller reached 87.6% success over 1000 episodes, versus 66.0% for vanilla RL and 9.30% for a central-pattern-generator torque baseline, and reached targets about 29% faster. Real-robot validation is left for future work, as the authors state.

What carries the argument

The load-bearing object is the Double Threshold Spiking (DTS) neuron, a single-neuron controller with no learned input weights: its discrete dynamics are $u_{k+1} = (1-|o_k|)(1-\frac{dt}{\tau})u_k + \frac{dt}{\tau}C_q q_k$, with output $o_k = 1$ if $u_k < u_n$, $o_k = -1$ if $u_k > u_p$, and $0$ otherwise, and torque $\Gamma_k = C_t o_k$. The two thresholds $u_n < u_p$ are the only parameters the reinforcement-learning agent adjusts. The mechanism it exploits is resonance-like: because the spike resets the membrane potential to zero and injects a torque impulse into the elastic segment, the controller can sustain and shape the body's own oscillation rather than imposing a motor trajectory. The same-sign threshold cases generate continuous spiking that bends a segment to a fixed posture, while opposite-sign thresholds confine deformation to a band, giving the controller a small action vocabulary from which gaits are composed.

What would settle it

Re-run the 1000-episode comparison after optimizing the settings of the oscillator-based baseline on this exact snake and reward (using the same genetic-algorithm procedure that produced the paper's settings); if its success rate rises to or above 87.6%, the claimed advantage over that baseline is an artifact of an undertuned comparison rather than a property of the spiking mechanism.

Watch

Extended reading notes

Core claim

The central claim is that the body of a soft snake is itself an oscillator, and a properly timed spiking controller can both excite and regulate that oscillation. The proposed Double Threshold Spiking (DTS) neuron is a leaky integrate-and-fire model whose membrane potential integrates the segment's deformation $d_k$; when the potential crosses an upper threshold it emits a negative torque spike, when it crosses a lower threshold it emits a positive torque spike, and otherwise it stays silent. These spikes act on the first node of the segment with an equal reaction torque on the last node. The paper shows in a mass-spring-damper example and in snake simulations that different threshold pairs drive the body to different stable oscillation patterns, so that forward motion and turning emerge from threshold choice alone. Paired with PPO, the high-level agent needs only to set $u_p = \mu + |\sigma|$ and $u_n = \mu - |\sigma|$ per segment, and the resulting controller achieves the reported success-rate and time improvements over torque-space RL and a CPG baseline.

Load-bearing premise

The load-bearing premise is that the comparison controller is a fair, representative version of the standard oscillator-based approach: its internal settings were taken from another study and not tuned for this task, so its near-random performance (9.30% success versus 9.70% for random actions) must reflect the approach itself rather than bad tuning; if that premise fails, the paper's claimed improvement over that baseline collapses.

Editorial extensions

If this is right

  • If the central claim holds, a soft snake can be steered by a very low-dimensional policy: instead of an RL agent emitting continuous torque commands, the agent emits two threshold parameters per segment and the DTS neuron converts them into gaits.
  • The controller scales across snake sizes without redesign: four simulated configurations (1, 3, and 5 segments and a 6-node version) all trained to high success rates between 79.3% and 91.3%.
  • The event-triggered outputs are silent (zero torque) 28.5% of time steps in the SpikingSoft tests, versus 13.5% for the CPG baseline, suggesting a direct route to energy savings by not driving the actuators continuously.
  • Zigzag forward motion and turning appear without any serpenoid curve or pre-programmed body-shape model, meaning the same controller could be reused for bodies whose optimal gaits are not known in advance.

Reading between the lines

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

  • Because the DTS neuron uses only the local deformation signal and has no learned weights, the same threshold-tuning scheme could plausibly transfer to other soft continuum robots whose elastic modes can be excited by discrete impulses; the paper does not test this.
  • A natural next experiment is to replace deformation with other oscillation metrics, such as strain, curvature, or velocity, and measure whether the same threshold-sharing structure still works, since the authors point to other oscillatory inputs as future work.
  • A sharper comparison would re-optimize the oscillator-based baseline's settings on this exact task and reward, to separate the benefit of the spiking mechanism itself from the benefit of task-specific tuning.
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

3 major / 6 minor

Summary. The paper proposes a Double Threshold Spiking (DTS) neuron as a low-level torque controller for each segment of a soft snake robot, with the neuron's two thresholds modulated by a PPO policy. The DTS neuron receives segment deformation as input and produces positive, negative, or zero torque spikes, leading to event-triggered actuation. The authors claim that this controller excites the natural oscillations of the soft body, enables turning and forward motion by threshold changes, and improves target-reaching performance compared with vanilla RL and a CPG controller acting in torque space. Experiments are carried out in a Cosserat-rod simulator with a 3-segment, 3-node soft snake, comparing SpikingSoft+PPO, vanilla PPO, CPG+PPO, and random actions. Reported results show SpikingSoft achieving 87.6% success rate versus 66.0% for vanilla PPO and 9.3% for the CPG baseline, with lower game time and higher total reward. The paper also introduces a 'Silence Rate' metric to quantify actuation sparsity.

Significance. If the central claims hold, the idea of using a single spiking neuron with two learnable thresholds as a torque-level gait generator for soft robots is a useful and interesting contribution. It connects spiking neural dynamics with exploitation of body elasticity, and the event-triggered nature of the controller could motivate energy-efficient control designs. The paper provides simulation results with means and standard deviations over five training runs and 1000 test episodes, which is a strength. However, the comparison with the CPG baseline is currently not convincing because that baseline performs at the level of random actions, and the 'smoother movements' claim is not quantitatively supported. The core mechanism claim about exciting natural oscillations also lacks direct analysis on the actual soft-snake model. These issues are fixable but need to be addressed before the comparative claims can be accepted.

major comments (3)
  1. [III, Table II] The CPG+PPO baseline reaches a 9.30% ± 3.13% success rate, which is statistically indistinguishable from the 9.70% ± 2.28% of the random-action baseline. The CPG hyperparameters (amplitude ratio 2.30, self/mutual inhibition weights 10.05 and 2.18, discharge rate 0.56, adaptation rate 1.76, coupling weights 9.13 and 0.73) are inherited from [19] rather than re-tuned for the snake geometry, target-radius-8 task, or reward function used here. Since PPO only adjusts the tonic inputs to the CPG oscillators, a CPG with mismatched intrinsic dynamics or an ill-scaled torque output can remain ineffective regardless of what PPO learns. Because the abstract explicitly claims improvement 'compared to ... Central Pattern Generator controller,' this baseline is load-bearing. Please re-optimize the CPG hyperparameters for this task, perform a sensitivity study, or explicitly drop the CPG comparison; otherwise the CPG-based performance claim is not established.
  2. [Abstract and Section IV] The claim of 'smoother movements' is not quantified anywhere in the paper. The only support is the qualitative after-image comparison in Fig. 8 and a sentence stating that the SpikingSoft snake 'tends to move forward with a smooth wiggle.' A smoothness metric should be defined and reported (for example, path curvature, jerk, torque variation, or deformation rate) in Table II or a dedicated table. Without such a metric, the smoothness claim is not testable and should either be removed or substantiated.
  3. [II-B and Fig. 2] The mechanism claim that the DTS neuron can 'excite the natural dynamics' or 'utilize natural oscillations' of the soft snake is demonstrated only on a linear mass-spring-damper in Fig. 2. For the actual Cosserat-rod soft snake, no frequency or energy analysis is provided to show that the observed oscillatory motion is related to a natural mode of the soft body rather than being an arbitrary bang-bang limit cycle. Please add an analysis of the snake segment's open-loop response, a comparison of spike frequency with the soft body's natural frequencies, or an energy input/transfer analysis. Alternatively, soften the claim to state that DTS produces periodic torque patterns that are effective with the body dynamics.
minor comments (6)
  1. [II-A] In the spike-generation definition, the function g(·) is called a 'Heaviside step function,' but it takes values 1, -1, and 0; please call it a sign/threshold function or define it explicitly.
  2. [III, Eq. (2)] The reward function r1 is written as '1 if l < 1; 5 else if l < 0.5; 10 else if l < 0.25.' With that ordering, the l < 0.5 and l < 0.25 branches are unreachable. Please adjust the inequalities or ordering if the intended reward is a graded step function that increases as the head approaches the target.
  3. [Table III] In the '3 node 3 segment' row, '26.42% + 7.57%' should read '26.42% ± 7.57%'.
  4. [Abstract] The phrase '21.6% increase in success rate' is ambiguous; because the comparison is 87.6% versus 66.0%, the correct statement is '21.6 percentage points' (or about 32.7% relative improvement).
  5. [III, Silence Rate] The Silence Rate is described as indicating 'greater energy efficiency,' but it only measures the fraction of zero output steps, not actual energy consumption. Report torque integral or power consumption if an energy claim is intended, or reframe the metric as actuation sparsity.
  6. [I, Introduction] The contribution statement 'We are the first to address the integration of bio-inspired spiking neurons as gait controllers ... of bionic soft robots' is somewhat overstated given the cited works [36]–[43] on spiking locomotion; please soften the novelty claim to focus on the specific soft-snake torque-level event-triggered mechanism.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the DTS/SpikingSoft derivation is self-contained and performance claims rest on simulation against external baselines.

full rationale

The paper's central derivation is self-contained. The DTS neuron is defined by its own update and spike-generation equations (Eq. 1), and the SpikingSoft controller is explicitly specified as taking the segment deformation as input and producing torque as output. The claim that this controller can excite natural oscillations is supported by the mass-spring-damper phase portraits and by the soft-snake simulations, not by reusing a fitted parameter as a prediction. The reinforcement-learning comparison is measured in simulation against vanilla PPO and a CPG baseline, and the reported success-rate and time improvements are empirical results, not consequences of the DTS equations by construction. The only author-overlapping citation is [31] (Wei Pan is a co-author), used in a related-work sentence about SNN-based RL; it is not load-bearing for the DTS model, the SpikingSoft architecture, or the performance claims. The CPG baseline hyperparameters are inherited from external work [19] and not re-tuned, which is a legitimate baseline-fairness concern but not a circularity issue. No step in the paper's derivation chain reduces to its own input by definition or by self-citation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central result depends on simulation fidelity, the chosen deformation feedback signal, and the fairness of baselines. Hand-set controller constants and reward weights are free parameters, and the CPG baseline inherits hyperparameters from prior work without re-tuning. The DTS neuron is a new construct with no independent evidence beyond the authors' simulations.

free parameters (5)
  • Cq (input gain) = 5.0
    Scales deformation input into the DTS neuron; hand-selected in Table I.
  • Ct (output torque gain) = 0.1
    Sets the magnitude of each torque pulse; hand-selected and not swept.
  • tau (membrane time constant) = 0.1
    Sets the DTS integration time; hand-selected in Table I.
  • Reward shaping coefficients = r2=250, r3=-l^2, r4=-5000
    Hand-designed in Eq. (2) to shape the learned policy toward the target.
  • CPG baseline hyperparameters = amplitude 2.30, weights 10.05/2.18, rates 0.56/1.76, coupling 9.13/0.73
    Inherited from [19] via genetic algorithm and not re-tuned for this snake task, which affects the fairness of the baseline comparison.
assumptions (4)
  • domain assumption The Cosserat rod simulation with Elastica accurately represents soft snake body dynamics, including bending, twisting, shear, and elongation.
    Section III and Table I specify model parameters; no hardware validation is provided.
  • domain assumption The deformation angle d, computed from node positions, is a sufficient feedback signal for the DTS controller to generate gaits.
    Section II-B defines d as the oscillation metric; alternative signals are not ablated.
  • domain assumption The DTS neuron produces stable oscillation in the full nonlinear snake body, analogous to the mass-spring-damper limit cycle in Fig. 2.
    Section II-A shows limit cycles only for a linear system; stability for the Cosserat rod is not proven.
  • domain assumption The anisotropic friction model matches snake-like ground contact.
    Table I lists friction coefficients; no experimental comparison or sensitivity analysis is provided.
invented entities (1)
  • Double Threshold Spiking (DTS) neuron
    purpose: Generates event-driven positive and negative torque pulses from deformation feedback and adjustable thresholds.
    Defined in Section II-A and evaluated only in the paper's own simulations; no independent physical measurement or falsifiable prediction outside the model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SpikingSoft: A Spiking Neuron Controller for Bio-inspired Locomotion with Soft Snake Robots." pith.science (2026). https://pith.science/paper/BHEHMK55

@misc{pith2026250119072,
  author       = {Pith},
  title        = {Pith review of: SpikingSoft: A Spiking Neuron Controller for Bio-inspired Locomotion with Soft Snake Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BHEHMK55}},
  note         = {Machine review of arXiv:2501.19072}
}
read the original abstract

Inspired by the dynamic coupling of moto-neurons and physical elasticity in animals, this work explores the possibility of generating locomotion gaits by utilizing physical oscillations in a soft snake by means of a low-level spiking neural mechanism. To achieve this goal, we introduce the Double Threshold Spiking neuron model with adjustable thresholds to generate varied output patterns. This neuron model can excite the natural dynamics of soft robotic snakes, and it enables distinct movements, such as turning or moving forward, by simply altering the neural thresholds. Finally, we demonstrate that our approach, termed SpikingSoft, naturally pairs and integrates with reinforcement learning. The high-level agent only needs to adjust the two thresholds to generate complex movement patterns, thus strongly simplifying the learning of reactive locomotion. Simulation results demonstrate that the proposed architecture significantly enhances the performance of the soft snake robot, enabling it to achieve target objectives with a 21.6% increase in success rate, a 29% reduction in time to reach the target, and smoother movements compared to the vanilla reinforcement learning controllers or Central Pattern Generator controller acting in torque space.

Figures

Figures reproduced from arXiv: 2501.19072 by the authors.

Figure 3
Figure 3. Example of DTS neuron dynamics. In this example, [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Example of the DTS neuron (left) and snake segment states with the SpikingSoft controller (right). [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Left: Example of the DTS neuron state when un and up are of the same sign. The figure shows three cases: 1) Green represents un equals 0.05 and up equals 0.075; 2) Blue means un equals 0.05 and up equals 0.15; 3) Red represents un equals 0.1 and up equals 0.15. Right: Three examples of the segment state with the SpikingSoft controller when un and up are of the same sign, which corresponds to left figure. same limit … view at source ↗
Figures from the paper (2 more)
Figure 7
Figure 7. Figure 7: Example of evaluation scenarios for the 3-segment 3-node snake with the SpikingSoft controller. The purple snake [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Test scenario performance comparison of the CPG, vanilla RL, and SpikingSoft controllers with MLP PPO. The first [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 40 canonical work pages

  1. [19]

    Reinforcement learning of cpg- regulated locomotion controller for a soft snake robot,

    X. Liu, C. D. Onal, and J. Fu, “Reinforcement learning of cpg- regulated locomotion controller for a soft snake robot,” IEEE Trans- actions on Robotics , 2023

    work page 2023

  2. [1]

    Deep- reinforcement-learning-based gait pattern controller on an uneven ter- rain for humanoid robots,

    P.-H. Kuo, C.-H. Pao, E.-Y . Chang, and H.-T. Yau, “Deep- reinforcement-learning-based gait pattern controller on an uneven ter- rain for humanoid robots,” International Journal of Optomechatronics, vol. 17, no. 1, p. 2222146, 2023

    work page 2023

  3. [2]

    A review on gait generation of the biped robot on various terrains,

    M. S. Khan and R. K. Mandava, “A review on gait generation of the biped robot on various terrains,” Robotica, vol. 41, no. 6, pp. 1888– 1930, 2023

    work page 1930

  4. [3]

    A novel pneumatic soft snake robot using traveling-wave locomotion in constrained environments,

    X. Qi, H. Shi, T. Pinto, and X. Tan, “A novel pneumatic soft snake robot using traveling-wave locomotion in constrained environments,” IEEE Robotics and Automation Letters , vol. 5, no. 2, pp. 1610–1617, 2020

    work page 2020

  5. [4]

    Snake robots for surgical applications: A review,

    J. Seetohul and M. Shafiee, “Snake robots for surgical applications: A review,” Robotics, vol. 11, no. 3, p. 57, 2022

    work page 2022

  6. [5]

    Review of snake robots in constrained en- vironments,

    J. Liu, Y . Tong, and J. Liu, “Review of snake robots in constrained en- vironments,” Robotics and Autonomous Systems , vol. 141, p. 103785, 2021

    work page 2021

  7. [6]

    Soft robots,

    C. Della Santina, M. G. Catalano, A. Bicchi, M. Ang, O. Khatib, and B. Siciliano, “Soft robots,” Encyclopedia of Robotics , vol. 489, 2020

    work page 2020

  8. [7]

    Developments and trend maps for soft continuum manipulators and soft snake robots,

    V . Rajashekhar, G. Prabhakar et al. , “Developments and trend maps for soft continuum manipulators and soft snake robots,” F oundations and Trends® in Robotics , vol. 12, no. 1, pp. 1–74, 2024

    work page 2024

Show all 46 references
  1. [8]

    Analysis of snake movement forms for realization of snake- like robots,

    S. Ma, “Analysis of snake movement forms for realization of snake- like robots,” in Proceedings 1999 IEEE International Conference on Robotics and Automation (Cat. No. 99CH36288C) , vol. 4. Ieee, 1999, pp. 3007–3013

    1999

  2. [9]

    Adaptive path following controller of a multijoint snake robot based on the improved serpenoid curve,

    D. Li, Z. Pan, H. Deng, and L. Hu, “Adaptive path following controller of a multijoint snake robot based on the improved serpenoid curve,” IEEE Transactions on Industrial Electronics , vol. 69, no. 4, pp. 3831– 3842, 2021

    2021

  3. [10]

    Soft robotic snake locomotion: Modeling and experimental assessment,

    D. D. Arachchige, Y . Chen, and I. S. Godage, “Soft robotic snake locomotion: Modeling and experimental assessment,” in 2021 IEEE 17th International Conference on Automation Science and Engineering (CASE). IEEE, 2021, pp. 805–810

    2021

  4. [11]

    A multigait continuous flexible snake robot for locomotion in complex terrain,

    W. Zhao, J. Wang, and Y . Fei, “A multigait continuous flexible snake robot for locomotion in complex terrain,” IEEE/ASME Transactions on Mechatronics, vol. 27, no. 5, pp. 3751–3761, 2021

    2021

  5. [12]

    Central pattern generators for locomotion control in animals and robots: a review,

    A. J. Ijspeert, “Central pattern generators for locomotion control in animals and robots: a review,” Neural networks , vol. 21, no. 4, pp. 642–653, 2008

    2008

  6. [13]

    Cpg-based control of serpentine locomotion of a snake-like robot,

    X. Wu and S. Ma, “Cpg-based control of serpentine locomotion of a snake-like robot,” Mechatronics, vol. 20, no. 2, pp. 326–334, 2010

    2010

  7. [14]

    A coach-based bayesian reinforcement learning method for snake robot control,

    Y . Jia and S. Ma, “A coach-based bayesian reinforcement learning method for snake robot control,” IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 2319–2326, 2021

    2021

  8. [15]

    An energy-saving snake locomotion pattern learned in a physically constrained environment with online model-based policy gradient method,

    Y . Liu and A. Barati Farimani, “An energy-saving snake locomotion pattern learned in a physically constrained environment with online model-based policy gradient method,” Journal of Mechanisms and Robotics, vol. 15, no. 4, p. 041007, 2023

    2023

  9. [16]

    Simulation to real: learning energy-efficient slithering gaits for a snake-like robot,

    Z. Bing, L. Cheng, K. Huang, and A. Knoll, “Simulation to real: learning energy-efficient slithering gaits for a snake-like robot,” IEEE Robotics & Automation Magazine , vol. 29, no. 4, pp. 92–103, 2022

    2022

  10. [17]

    Deep reinforcement learning for snake robot locomotion,

    J. Shi, T. Dear, and S. D. Kelly, “Deep reinforcement learning for snake robot locomotion,” IF AC-PapersOnLine, vol. 53, no. 2, pp. 9688–9695, 2020

    2020

  11. [18]

    Reinforce- ment learning of serpentine locomotion for a snake robot,

    K. Qiu, H. Zhang, Y . Lv, Y . Wang, C. Zhou, and R. Xiong, “Reinforce- ment learning of serpentine locomotion for a snake robot,” in 2021 IEEE International Conference on Real-time Computing and Robotics (RCAR). IEEE, 2021, pp. 468–473

    2021

  12. [20]

    Cpg-rl: Learning central pattern generators for quadruped locomotion,

    G. Bellegarda and A. Ijspeert, “Cpg-rl: Learning central pattern generators for quadruped locomotion,” IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 12 547–12 554, 2022

    2022

  13. [21]

    Manyquadrupeds: Learn- ing a single locomotion policy for diverse quadruped robots,

    M. Shafiee, G. Bellegarda, and A. Ijspeert, “Manyquadrupeds: Learn- ing a single locomotion policy for diverse quadruped robots,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 3471–3477

    2024

  14. [22]

    Motor control: Snake neurons speed up,

    S. Bello-Rojas and M. W. Bagnall, “Motor control: Snake neurons speed up,” Current Biology, vol. 34, no. 3, pp. R98–R99, 2024

    2024

  15. [23]

    Opportunities for neuromorphic computing algorithms and applications,

    C. D. Schuman, S. R. Kulkarni, M. Parsa, J. P. Mitchell, B. Kay et al. , “Opportunities for neuromorphic computing algorithms and applications,” Nature Computational Science , vol. 2, no. 1, pp. 10– 19, 2022

    2022

  16. [24]

    A cerebellar-based solution to the nondeterministic time delay problem in robotic control,

    I. Abad ´ıa, F. Naveros, E. Ros, R. R. Carrillo, and N. R. Luque, “A cerebellar-based solution to the nondeterministic time delay problem in robotic control,” Science Robotics, vol. 6, no. 58, p. eabf2756, 2021

    2021

  17. [25]

    Openai gym,

    G. Brockman, V . Cheung, L. Pettersson, J. Schneider, J. Schul- man, J. Tang, and W. Zaremba, “Openai gym,” arXiv preprint arXiv:1606.01540, 2016

    2016 arXiv

  18. [26]

    Deep reinforcement learning with spiking q-learning,

    D. Chen, P. Peng, T. Huang, and Y . Tian, “Deep reinforcement learning with spiking q-learning,” arXiv preprint arXiv:2201.09754 , 2022

    2022 arXiv

  19. [27]

    Porting deep spiking q-networks to neuromorphic chip loihi,

    M. Akl, Y . Sandamirskaya, F. Walter, and A. Knoll, “Porting deep spiking q-networks to neuromorphic chip loihi,” in International Conference on Neuromorphic Systems 2021 , 2021, pp. 1–7

    2021

  20. [28]

    Deep reinforcement learning with population-coded spiking neural network for continuous control,

    G. Tang, N. Kumar, R. Yoo, and K. Michmizos, “Deep reinforcement learning with population-coded spiking neural network for continuous control,” in Conference on Robot Learning . PMLR, 2021, pp. 2016– 2029

    2021

  21. [29]

    Fully spiking actor network with intralayer connections for reinforcement learning,

    D. Chen, P. Peng, T. Huang, and Y . Tian, “Fully spiking actor network with intralayer connections for reinforcement learning,” IEEE Transactions on Neural Networks and Learning Systems , 2024

    2024

  22. [30]

    End to end learning of spiking neural network based on r- stdp for a lane keeping vehicle,

    Z. Bing, C. Meschede, K. Huang, G. Chen, F. Rohrbein, M. Akl, and A. Knoll, “End to end learning of spiking neural network based on r- stdp for a lane keeping vehicle,” in2018 IEEE international conference on robotics and automation (ICRA) . IEEE, 2018, pp. 4725–4732

    2018

  23. [31]

    Spiking neural-networks-based data-driven con- trol,

    Y . Liu and W. Pan, “Spiking neural-networks-based data-driven con- trol,” Electronics, vol. 12, no. 2, p. 310, 2023

    2023

  24. [32]

    Target tracking control of a wheel-less snake robot based on a supervised multi-layered snn,

    Z. Jiang, R. Otto, Z. Bing, K. Huang, and A. Knoll, “Target tracking control of a wheel-less snake robot based on a supervised multi-layered snn,” in 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2020, pp. 7124–7130

    2020

  25. [33]

    Biologically inspired dynamic thresholds for spiking neural net- works,

    J. Ding, B. Dong, F. Heide, Y . Ding, Y . Zhou, B. Yin, and X. Yang, “Biologically inspired dynamic thresholds for spiking neural net- works,” Advances in Neural Information Processing Systems , vol. 35, pp. 6090–6103, 2022

    2022

  26. [34]

    Low latency and sparse computing spiking neural networks with self-driven adaptive threshold plasticity,

    A. Zhang, J. Shi, J. Wu, Y . Zhou, and W. Yu, “Low latency and sparse computing spiking neural networks with self-driven adaptive threshold plasticity,” IEEE Transactions on Neural Networks and Learning Systems , 2023

    2023

  27. [35]

    Construct- ing accurate and efficient deep spiking neural networks with double- threshold and augmented schemes,

    Q. Yu, C. Ma, S. Song, G. Zhang, J. Dang, and K. C. Tan, “Construct- ing accurate and efficient deep spiking neural networks with double- threshold and augmented schemes,” IEEE Transactions on Neural Networks and Learning Systems , vol. 33, no. 4, pp. 1714–1726, 2021

    2021

  28. [36]

    A cpg system based on spiking neurons for hexapod robot locomotion,

    H. Rostro-Gonzalez, P. A. Cerna-Garcia, G. Trejo-Caballero, C. H. Garcia-Capulin, M. A. Ibarra-Manzano, J. G. Avina-Cervantes, and C. Torres-Huitzil, “A cpg system based on spiking neurons for hexapod robot locomotion,” Neurocomputing, vol. 170, pp. 47–54, 2015

    2015

  29. [37]

    Combining spiking motor primitives with a behaviour-based architecture to model locomotion for six-legged robots,

    J. C. V . Tieck, J. Rutschke, J. Kaiser, M. Schulze, T. Buettner, D. Reichard, A. Roennau, and R. Dillmann, “Combining spiking motor primitives with a behaviour-based architecture to model locomotion for six-legged robots,” in 2019 IEEE/RSJ International Conference on Intellig...

    2019

  30. [38]

    Locomotion control of robotic fish with a hierarchical framework combining spiking neural networks and cpgs,

    M. Wang, X. Li, Y . Zhang, Z. Chang, and J. Yu, “Locomotion control of robotic fish with a hierarchical framework combining spiking neural networks and cpgs,” in 2019 IEEE 9th Annual International Confer- ence on CYBER Technology in Automation, Control, and Intelligent Systems...

    2019

  31. [39]

    Learning to walk: Bio-mimetic hexapod locomotion via reinforcement-based spiking central pattern generation,

    A. S. Lele, Y . Fang, J. Ting, and A. Raychowdhury, “Learning to walk: Bio-mimetic hexapod locomotion via reinforcement-based spiking central pattern generation,” IEEE Journal on Emerging and Selected Topics in Circuits and Systems , vol. 10, no. 4, pp. 536–545, 2020

    2020

  32. [40]

    An end-to-end spik- ing neural network platform for edge robotics: From event-cameras to central pattern generation,

    A. Lele, Y . Fang, J. Ting, and A. Raychowdhury, “An end-to-end spik- ing neural network platform for edge robotics: From event-cameras to central pattern generation,” IEEE Transactions on Cognitive and Developmental Systems , vol. 14, no. 3, pp. 1092–1103, 2021

    2021

  33. [41]

    Spiking neural networks and their applications: A review,

    K. Yamazaki, V .-K. V o-Ho, D. Bulsara, and N. Le, “Spiking neural networks and their applications: A review,” Brain Sciences , vol. 12, no. 7, p. 863, 2022

    2022

  34. [42]

    Fully spiking neural network for legged robots,

    X. Jiang, Q. Zhang, J. Sun, and R. Xu, “Fully spiking neural network for legged robots,” arXiv preprint arXiv:2310.05022 , 2023

    2023 arXiv

  35. [43]

    Toward brain-inspired learning with the neuromorphic snake-like robot and the neurorobotic platform,

    G. Chen, Z. Bing, F. R ¨ohrbein, J. Conradt, K. Huang, L. Cheng, Z. Jiang, and A. Knoll, “Toward brain-inspired learning with the neuromorphic snake-like robot and the neurorobotic platform,” IEEE Transactions on Cognitive and Developmental Systems , vol. 11, no. 1, pp. 1–12, 2017

    2017

  36. [44]

    Spatio-temporal backpropagation for training high-performance spiking neural networks,

    Y . Wu, L. Deng, G. Li, and L. Shi, “Spatio-temporal backpropagation for training high-performance spiking neural networks,” Frontiers in neuroscience, vol. 12, p. 323875, 2018

    2018

  37. [45]

    Modeling and simulation of complex dynamic musculoskeletal architectures,

    X. Zhang, F. Chan, T. Parthasarathy, and M. Gazzola, “Modeling and simulation of complex dynamic musculoskeletal architectures,” Nature Communications , vol. 10, no. 1, pp. 1–12, 2019. [Online]. Available: https://doi.org/10.1038/s41467-019-12759-5

    2019 doi

  38. [46]

    Proximal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.