Pith. sign in

REVIEW 3 major objections 4 minor 27 references

Ace! Motion Planning of Professional-Level Table Tennis Serves with a Robot Arm

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Pith's one-line read A robot arm can serve table tennis at professional level — spins up to 550 rad/s and speeds up to 6.7 m/s — while obeying official ITTF rules, according to this motion-planning framework.

desk verdict Real ITTF-legal professional-level serves from a robot arm, genuinely new, but the 'surpassing elite' headline overstates a pruned and curated subset. read the letter →

arxiv 2607.06989 v2 pith:TR3H24AW submitted 2026-07-08 cs.RO cs.SYeess.SY

classification cs.ROcs.SYeess.SY MSC 68T40
keywords tabletennisrobotservemotionplanningmodelpredictivecontrolBayesianoptimizationITTFrulesspingenerationsim-to-real
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 robot arm can serve in table tennis at a professional level: generating high spin (up to 550 rad/s) and speed (up to 6.7 m/s) from a spinless free-falling ball, while remaining fully compliant with ITTF service rules. The authors argue their pipeline — a parametrized minimum-jerk motion planner plus a Bayesian optimizer that searches over racket contact state in simulation — is the first to achieve this. In officiated matches against elite and professional players, the robot's serves won points directly (aces) in 8 to 23 percent of its serves, and serving win probabilities reached 51.5–58.7%, slightly above published elite human stats. The claim matters because the serve is the only fully controllable action in table tennis, and previous robot serve work focused on placement without high spin or a legal toss. The paper rests on zero-shot open-loop execution: plans are trained in simulation and deployed without real-world adaptation, with a conceded sim-to-real gap that reduces transfer efficiency.

What carries the argument

The key machinery is a two-stage pipeline: (1) a parametrized model-predictive motion planner that solves a constrained minimum-jerk optimization (Eq. 2) to move the racket to a specified pose, orientation, and velocity at a specified time, then return smoothly to rest; and (2) HEBO, a heteroscedastic and evolutionary Bayesian optimizer, which searches over up to ten normalized parameters — hit timing, racket velocity vector, racket yaw/roll orientation, return duration, and optional position offset — using a simulator rollout with a legality-check chain to score each candidate. The racket's desired intercept point is tied to the expected tossed-ball trajectory, so the optimizer effectively

What would settle it

Run the competition library's serves on the real robot and measure the ball's linear and angular velocity at the second bounce with an independent multi-camera tracking system: if mean spin falls below the reported task ranges (e.g., below 200 rad/s for the topspin task) or if more than 20% of serves fail umpire legality review in a fresh session, the central claim would be contradicted.

Watch

Extended reading notes

Core claim

The central discovery is that by combining a hybrid end-effector/joint-space minimum-jerk motion planner with a sample-efficient Bayesian optimizer (HEBO), the robot can find racket contact states (time, velocity, orientation, return duration) that turn a tossed ball into a legal, hard-to-return serve. The optimizer rolls each candidate plan out in a physics simulator, applies a sequential chain of ITTF legality checks, and rewards spin, velocity, placement, and low trajectory height. The resulting serve library, pruned by real tournament outcomes, produced measured spin and velocity that the authors report match or exceed those of professional players, and that won direct points against Oly

Load-bearing premise

The whole serve library is planned against a physics simulator and executed open-loop, so everything rests on the simulator's ball-flight, bounce, and impact models being accurate enough that ITTF legality and high spin/speed carry over to the real table, and on the N=15 recorded toss trajectories remaining representative of future tosses.

Editorial extensions

If this is right

  • If the claim holds, robot table tennis can now include a professional-caliber serve, closing the last fully 'human' phase of the game.
  • The legality-check chain in simulation plus open-loop deployment suggests that simulation-based training can produce rule-compliant behaviors without real-world trial and error, at least for compact, high-speed sports actions.
  • Because the same motion planner handles the toss and the strike, the framework extends to other tasks requiring precise, high-speed interception with secondary objectives (e.g., placement or spin).
  • The reported serving win probabilities (51.5–58.7%) slightly exceed published elite male/female averages, implying the serve is at least as effective as a human pro's serve.
  • The zero-shot transfer rate (12–15 of 25 plans per task) defines a concrete ceiling: a quarter to half of simulated serves still need to be filtered out by real-world trials.

Reading between the lines

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

  • One testable extension would be to add a per-toss closed-loop adjustment of hit timing based on online ball-tracking; if the reported ace rates are driven partly by toss prediction errors, closing the loop should raise the zero-shot transfer rate well above the current 48–60%.
  • The reported ace rates vary from 1.6% to 21% across tournaments, and the authors attribute gains to pruning and spin/topspin prioritization; this suggests the library composition, not the motion planner alone, carries much of the matchup effectiveness — a hypothesis that could be checked by ablating the pruning rule.
  • The abstract's headline spin of 550 rad/s is an optimized extreme, while task-specific tables show lower, more variable typical values (e.g., topspin task real spin 263±70 rad/s); readers should distinguish the claimed ceiling from the library's typical performance.
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 / 4 minor

Summary. The paper presents a framework for generating ITTF-compliant table tennis serves with an 8-DoF robot arm. The pipeline combines a jerk-minimizing Model Predictive Control motion planner with the HEBO Bayesian optimizer to search over racket strike parameters in simulation, followed by offline filtering and iterative real-world selection into a competition library. The authors claim that this framework produces serves with spin up to 550 rad/s and speeds up to 6.7 m/s, matching or surpassing elite human players, and report match statistics against professional players (including Olympian Miu Hirano) with ace rates up to 21%. The central claim is that this is the first framework capable of generating professional-level, ITTF-compliant serve motion plans for a robot arm.

Significance. If the claim is substantiated, this is a significant advance in robot table tennis and dynamic manipulation. The real-world validation is unusually strong for a robotics paper: licensed umpires, professional opponents, and counted direct serve points. The combination of an optimization-based motion planner with Bayesian-optimized strike parameters is a technically plausible and potentially generalizable approach. The paper also provides quantitative data on spin, velocity, and aiming error, which is valuable to the community. However, the headline claim of 'matching and surpassing elite players' is weakened by selection effects in the reported data, as detailed in the major comments.

major comments (3)
  1. [Section IV-A, Table III] The real-world spin/velocity statistics and the comparison to elite players are computed only over the subset of serves that transferred zero-shot (N=12-15 of 25). The paper states: 'The statistics reported in the real world are thus based on the subset of serves that actually transferred zero-shot without issues.' This selection bias makes the 'matching and even surpassing elite players' claim an artifact of curation rather than a property of the generative framework. Please report the yield, legality rate, and full distributions for all 25 generated plans, or explicitly restrict the claim to the curated library.
  2. [Section III-E, Table II] The improvement narrative in Table II confounds algorithmic pipeline changes with strategic library pruning. The March and April 2026 rows explicitly include 'strategically prune and replace serves whose historical performance shows a tendency to lead to opponent's points.' This means the increase in ace rate could be due solely to the removal of poor serves, not to the HEBO, toss-robustness, or dynamic-limit changes. To support the claim that the pipeline improves, provide an ablation that separates the effect of library curation from generation-side improvements.
  3. [Section IV-B, Eqs. (3)-(4)] The statement 'strong correlation between the serve property encouraged through the reward function and the corresponding physical metric' is circular. The reward functions in Eqs. (3)-(4) directly include the physical metric being measured (e.g., r_top maximizes topspin). Finding that topspin-rewarded runs produce higher topspin is a sanity check, not a validation of the framework. Please rephrase or provide a non-circular validation, such as comparing against a baseline without the corresponding reward term.
minor comments (4)
  1. [Abstract, Section IV-A] The abstract claims spin 'up to 550 rad/s' but I could not find a table or figure in the provided text that reports this maximum. Please cite the specific figure/table that supports the 550 rad/s value.
  2. [Section III-C, Eq. (2)] The text mentions 'n_l = 323rd degree order polynomials' and later 'n_l = 32'. This appears to be a formatting error; likely it should read 'n_l = 32 third-degree polynomials.' Please clarify.
  3. [Section IV-A, Table II] The claim of 'slightly higher winning probabilities while serving' compared to the elite human statistics from [27] should be qualified: the 95% Wilson confidence intervals in April 2026 (51.5-58.7%) overlap with the elite values (52.78% and 53.28%). A formal statistical comparison would strengthen the claim.
  4. [References] The paper relies heavily on the authors' own prior Nature paper [1] as the 'baseline method.' Please explicitly state the novel contributions of this manuscript relative to [1] to avoid any ambiguity about the incremental advance.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-definitional validation; central serve-generation claim is independent of its inputs.

  1. self definitional [Section IV-B (Task-specific serve generation example), after Table III; reward definitions in Section III-D, Eqs. (3)-(4).]
    "The results indeed suggest that there is a strong correlation between the serve property encouraged through the reward function and the corresponding physical metric (in bold font). For example, when topspin is rewarded, the average value will become significantly larger for both Sim and Real setups (218 to 263rad/s) but the sign is flipped when backspin is encouraged instead (-131 to -164rad/s)."

    The reward terms in Eqs. (3)-(4) are defined as monotone functions of the same physical spin components reported in Table III: r_top = g_top max(0, ω_tb)/ω_bar and r_side = g_side φ(ω_ss). Thus the observed 'strong correlation' between the property encouraged by the reward and the measured spin metric is guaranteed by the objective's definition; it is a sanity check that the optimizer follows its reward, not an independent empirical confirmation. The step is local and does not support the central professional-level claim, so it is minor rather than load-bearing.

full rationale

The central pipeline is not circular: HEBO optimizes motion-plan parameters against a simulator using a reward constructed from explicit spin/velocity/placement targets, and the reported real spin/speed values are output measurements, not fitted inputs or predictions of those inputs. The 'up to 550 rad/s' claim is a measured capability, not an input to the method. The only genuinely definitional step is the Section IV-B sanity check that rewarded spin components correlate with the measured spin components; because Eqs. (3)-(4) define r_top/r_back/r_side as functions of the same ω_tb/ω_ss tabulated in Table III, that correlation follows by construction. This does not feed into the central claim and is not load-bearing. Self-citation to [1] appears only as a baseline, not as evidence for the present method. Concerns about the pruned/curated competition library and the subset of zero-shot-transferred serves are real external-validity/selection limitations, but they are not circularity under the definitions used here.

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

All numbers behind the headline spin/speed claims trace to hand-chosen engineering degrees of freedom: reward gains (§III-D) select which serve types are discovered; slack weights/tolerances (Table I) and legality-check penalties (§III-D) define what counts as feasible and legal; HEBO genome bounds (§III-D) cap the achievable velocity/orientation. The physics models (ball flight with Magnus effect, racket–ball impact, table bounce) are treated as given and are not specified numerically in the text; their fidelity is the load-bearing link between simulation-optimized plans and real, umpire-verified serves. The toss distribution (Fig. 3, N=15) is an empirical input whose stationarity is assumed for open-loop execution. No new physical entities are invented. These are normal engineering parameters rather than signs of fraud, but they mean the headline results cannot be separated from manual configuration choices.

free parameters (5)
  • Serve-reward gains (g_back, g_top, g_side; weights for r_pos, r_vel, r_height) = illustrative only: g_back=1, g_top=0, g_side=1, d=-1 (§III-D)
    Hand-chosen per task; they define which serve types the optimizer discovers, so the reported spin/speed distributions depend on them.
  • MPC slack weights and tolerances = λ_p=2 m⁻¹, λ_n=1, λ_v=0.2 (m/s)⁻¹; Δp=5 mm, Δφ=0.5°, Δv_a=0.01 m/s, Δv_r=0.01 (Table I)
    Hand-chosen; they set how accurately the racket hits the state HEBO requests, directly bounding achievable spin and speed.
  • Legality-chain penalties s_leg and MPC infeasibility penalty = P_MPC=-10; s_leg ∈ {-5,-4,-3,-2.5,-2,-1.5,-0.5,-0.3,-0.1} (§III-D)
    Hand-chosen shaping terms; they determine how the optimizer trades off legality violations and which serves are ever accepted.
  • HEBO genome bounds and hyperparameters = v_τ∈[±10,±5,±5] m/s, δp∈[±0.07]³ m, yaw/roll [π/2±π, ±π]; pop 20, 100 iterations
    Hand-chosen search domain caps the achievable velocities/orientations and hence the headline 550 rad/s and 6.7 m/s values.
  • Spin normalization clamp ω̄ = not numerically specified in text
    Used to normalize spin rewards in Eq. (4); its value shapes the reward landscape but is never reported, impeding replication.
assumptions (5)
  • domain assumption Ball-flight, racket–ball impact, and table-bounce simulation models are accurate enough that simulation-optimized serve plans transfer to reality
    Invoked in §III-D, where every HEBO candidate is rolled out in simulation and rewarded only if legal; Table III shows only 12–15 of 25 plans per task transferred zero-shot, and §V concedes the sim-to-real gap.
  • domain assumption Joint dynamics are linear, second-order, and decoupled across the 8 joints (model f_D)
    Stated in §III-A: "under the assumption that all joints can be modeled independently from the rest"; the polytopic constraint set P and all feasibility constraints are built from this model.
  • domain assumption The N=15 recorded toss trajectories (Fig. 3) represent deployment-time toss statistics
    Robustness training samples from these recordings (§III-D); serves are executed open-loop (§IV-B), so toss stationarity under air flow, vibration, and ball wear is load-bearing.
  • domain assumption The sequential legality check chain correctly encodes ITTF serve rules
    §III-D lists the checks (contact with non-racket body, bounce counts/sides, net clearance, edge distance). Real umpires at tournaments provide the ultimate external check.
  • domain assumption Optimization-based re-targeting of human toss demonstrations preserves a toss consistent with the simulator's assumptions
    §III-B re-targets human one-arm tosses to robot kinematics via [23]; validity is established only empirically by the ±1 cm spread in Fig. 3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ace! Motion Planning of Professional-Level Table Tennis Serves with a Robot Arm." pith.science (2026). https://pith.science/paper/TR3H24AW

@misc{pith2026260706989,
  author       = {Pith},
  title        = {Pith review of: Ace! Motion Planning of Professional-Level Table Tennis Serves with a Robot Arm},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TR3H24AW}},
  note         = {Machine review of arXiv:2607.06989}
}
read the original abstract

Table tennis, a dynamic, compact, and popular sport, has received significant attention as a robotics benchmark over the last decades. Most of the research has focused on the rally aspect - returning an incoming ball - requiring high-speed vision, agile motion planning, and tight closed-loop control. However, the other component of table tennis gameplay - the serve - is comparatively a quite unexplored research problem, that in fact requires pushing physics modeling and control to the extremes. Achieving competitive serves with a robot presents domain-specific challenges, such as high-spin generation from a spinless ball, precise aiming, or multi-objective optimization. In this work, we present a novel approach for generating official rule-compliant serves by combining motion primitives, Model Predictive Control, and Bayesian Optimization. Serves generated in this way offer a wide and controllable variation of spins of up to 550 rad/s, and speeds of up to 6.7 m/s, matching and even surpassing those of elite table tennis players.

Figures

Figures reproduced from arXiv: 2607.06989 by the authors.

Figure 1
Figure 1. A high side-spin serve executed by the robot. The ball trajectory [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Our robot arm holding a ball on the ball cup, about to serve. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. (a) Real ball trajectories of serves used during the match against the Paris 2024 Olympic silver medalist Miu Hirano. (b) Ball velocities and heights as [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 1 linked inside Pith

  1. [15]

    Learning to serve: An experimen- tal study for a new learning from demonstrations frame- work,

    O. Koç and J. Peters, “Learning to serve: An experimen- tal study for a new learning from demonstrations frame- work,”IEEE Robotics and Automation Letters, vol. 4, no. 2, pp. 1784–1791, 2019

  2. [16]

    Seal: A sample-efficient adjustment-learning method for table tennis robot serve,

    Q. Guo, X. Shi, K. Murakami, R. Jia, and Y . Yamakawa, “Seal: A sample-efficient adjustment-learning method for table tennis robot serve,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 14 924–14 930

  3. [17]

    Ball trajectory planning in serving task for table tennis robot,

    Y . Hayakawa, A. Nakashima, S. Itoh, and Y . Nakai, “Ball trajectory planning in serving task for table tennis robot,” SICE Journal of Control, Measurement, and System Integration, vol. 9, pp. 50–59, 2016

  4. [1]

    Outplaying elite table tennis players with an autonomous robot,

    P. Dürret al., “Outplaying elite table tennis players with an autonomous robot,”Nature, vol. 652, no. 8111, pp. 886–891, Apr 2026

  5. [2]

    "speed and spin characteristics of the 40mm table tennis ball

    H.-P. Tang, M. Mizoguchi, and S. Toyoshima, “"speed and spin characteristics of the 40mm table tennis ball",” Table tennis sciences, vol. 4, pp. 278–284, 2002

  6. [3]

    Learn- ing to select and generalize striking movements in robot table tennis,

    K. Mülling, J. Kober, O. Kroemer, and J. Peters, “Learn- ing to select and generalize striking movements in robot table tennis,”The International Journal of Robotics Re- search, vol. 32, no. 3, pp. 263–279, 2013

  7. [4]

    A robot ping-pong player: Experiment in real-time intelligent control,

    A. Russell, “A robot ping-pong player: Experiment in real-time intelligent control,” inAT Bell Laboratories. MIT Press, 1988

  8. [5]

    Learning to play table tennis from scratch using muscular robots,

    D. Büchler, S. Guist, R. Calandra, V . Berenz, B. Schölkopf, and J. Peters, “Learning to play table tennis from scratch using muscular robots,”IEEE Transactions on Robotics, vol. 38, no. 6, pp. 3850–3860, 2022

Show all 27 references
  1. [6]

    Achieving human level com- petitive robot table tennis,

    D. B. D’Ambrosioet al., “Achieving human level com- petitive robot table tennis,” in7th Robot Learning Work- shop: Towards Robots with Human-Level Abilities, 2025

  2. [7]

    Muelling, A

    K. Muelling, A. Boularias, B. Mohler, B. Schölkopf, and J. Peters, “Learning strategies in table tennis using 8 TABLE III SERVE STATISTICS IN SIMULATION AND IN THE REAL WORLD FOR DIFFERENT TASK SETTINGS Task Environment Aim error [mm] Velocity [m/s] Spin [rad/s] x-y norm x y z...

  3. [8]

    Robotic table tennis: A case study into a high speed learning system,

    D. B. D’Ambrosioet al., “Robotic table tennis: A case study into a high speed learning system,”arXiv preprint arXiv:2309.03315, 2023

  4. [9]

    The ping pong robot to return a ball precisely trajectory predic- tion and racket control for spinning balls,

    A. Kyohei, N. Masamune, and Y . Satoshi, “The ping pong robot to return a ball precisely trajectory predic- tion and racket control for spinning balls,” inOMRON Technics, vol. 51, no. 016, 2019

  5. [10]

    Analysis of the spatial distribution of the serve and the type of serve-return in elite table tennis. Sex differences,

    F. Pradas, V . Toro-Román, C. Castellar, and L. Carrasco, “Analysis of the spatial distribution of the serve and the type of serve-return in elite table tennis. Sex differences,” Physical Education Theory and Methodology, 2023

  6. [11]

    Serve analyses of elite european table tennis matches,

    Z. Djokic, “Serve analyses of elite european table tennis matches,”International Journal of Racket Sports Science, 01 2020

  7. [12]

    Com- parison of rally time in xxix beijing (2008) and xxviii athens (2004) olympic table tennis tournaments,

    M. Katsikadelis, T. Pilianidis, and A. Misichroni, “Com- parison of rally time in xxix beijing (2008) and xxviii athens (2004) olympic table tennis tournaments,”Int. J. Table Tennis Sci, vol. 6, pp. 55–59, 2010

  8. [13]

    International ta- ble tennis federation (ittf),

    International Table Tennis Federation, “International ta- ble tennis federation (ittf),” At: https://www.ittf.com, 2025

  9. [14]

    Hebo: Pushing the limits of sample-efficient hyperparameter optimisation,

    A. Cowen-Riverset al., “Hebo: Pushing the limits of sample-efficient hyperparameter optimisation,”Journal of Artificial Intelligence Research, vol. 74, 07 2022

  10. [18]

    High speed robotic table tennis swinging using lightweight hardware with model predictive control,

    D. Nguyen, K. D. Cancio, and S. Kim, “High speed robotic table tennis swinging using lightweight hardware with model predictive control,” in2025 IEEE Interna- tional Conference on Robotics and Automation (ICRA), 2025, pp. 15 278–15 284

  11. [19]

    Catching objects in flight,

    S. Kim, A. Shukla, and A. Billard, “Catching objects in flight,”IEEE Transactions on Robotics, vol. 30, no. 5, pp. 1049–1065, 2014

  12. [20]

    Agile catching with whole-body mpc and blackbox policy learning,

    S. Abeyruwanet al., “Agile catching with whole-body mpc and blackbox policy learning,” inProceedings of The 5th Annual Learning for Dynamics and Control Con- ference, ser. Proceedings of Machine Learning Research, N. Matni, M. Morari, and G. J. Pappas, Eds., vol. 211. PMLR, 15...

  13. [21]

    Learning-based design and control for quadrupedal robots with parallel-elastic actuators,

    F. Bjelonic, J. Lee, P. Arm, D. Sako, D. Tateo, J. Peters, and M. Hutter, “Learning-based design and control for quadrupedal robots with parallel-elastic actuators,”IEEE Robotics and Automation Letters, vol. 8, no. 3, pp. 1611– 1618, 2023

  14. [22]

    Toward real-world automated anti- body design with combinatorial bayesian optimization,

    M. A. Khanet al., “Toward real-world automated anti- body design with combinatorial bayesian optimization,” Cell Reports Methods, vol. 3, no. 1, p. 100374, 2023

  15. [23]

    Ac- quiring and generalizing the embodiment mapping from human observations to robot skills,

    G. Maeda, M. Ewerton, D. Koert, and J. Peters, “Ac- quiring and generalizing the embodiment mapping from human observations to robot skills,”IEEE Robotics and Automation Letters, vol. 1, no. 2, pp. 784–791, 2016

  16. [24]

    Smoothing and differ- entiation of data by simplified least squares procedures

    A. Savitzky and M. J. E. Golay, “Smoothing and differ- entiation of data by simplified least squares procedures.” Analytical Chemistry, vol. 36, no. 8, pp. 1627–1639, 1964

  17. [25]

    Coal: an extension of the flexible collision library,

    J. Pan, S. Chitta, D. Manocha, F. Lamiraux, J. Mirabel, J. Carpentier, L. Montautet al., “Coal: an extension of the flexible collision library,” https://github.com/coal- library/coal, 2015–2024

  18. [26]

    R. H. Byrd, J. Nocedal, and R. A. Waltz,Knitro: An In- tegrated Package for Nonlinear Optimization, G. Di Pillo and M. Roma, Eds. Boston, MA: Springer US, 2006

  19. [27]

    Analysis of contextual-related variables on serve and receiving performances in elite men’s and women’s table tennis players,

    M.-Á. Gómez, A. García-de Alcaráz, and P. Furley, “Analysis of contextual-related variables on serve and receiving performances in elite men’s and women’s table tennis players,”International Journal of Performance Analysis in Sport, vol. 17, no. 6, pp. 919–933, 2017

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.