Source-linked AI summary
Multi-bounce Drum Roll with Optimized Active Tricks to Leverage Soft Embodiment
Naoto Yamanaka, Takanori Jin, Taisuke Kobayashi
TL;DR
High-frequency drum rolls require multiple rebounds from one stroke, but robotic reproduction must coordinate elasticity with fine active motion. The paper introduces Tap-Pull and Micro-Pulse tricks and tunes them with Bayesian optimization. Optimized soft embodiments achieved over 12 bounces and 6.8-times higher acoustic efficiency than the rigid comparison.
Problem
Robotic drum rolls must produce rapid, uniform rebounds, while passive elasticity cannot fully shape sound and real-world optimization is limited by sample efficiency.
Method
The paper combines soft elastic end-effectors with Tap-Pull and Micro-Pulse active motions, tuning their parameters independently using Bayesian optimization and acoustic objectives.
Results
Optimized tricks increased bounce count to over 12 and improved acoustic efficiency by 6.8 times, with soft embodiment amplifying parameter effects.
Takeaways & Limitations
Active tricks combined with elasticity and optimization successfully produced sustained, uniform, and efficient drum rolls.
Takeaways & Limitations
Experiments used a fixed drum configuration and limited tempo range, and perceptual drum-roll quality was not explicitly evaluated.
Abstract
from arXiv · showhide
This paper presents a soft robotic drummer for accurate and efficient drum rolls. High-frequency drum rolls require the "multi-bounce technique," where a drumstick bounces multiple times with a single stroke. In robotic reproduction of this technique, the body's elasticity is key, while the fine motion during the stroke is also crucial for maximizing the potential of that elasticity. Therefore, we design two tricks: i) Tap-Pull (TP) trick to increase the number of rebounds by adding a pulling motion after impact; and ii) Micro-Pulse (MP) trick to keep the drumming volume by injecting small oscillations during the stroke. Due to the nonlinear complexity of soft embodiment, both tricks are efficiently tuned using Bayesian optimization in a data-driven manner for accomplishing the respective objectives quantified. We evaluated the optimized behaviors with soft and rigid end-effectors. As a result, the soft TP achieved the highest bounce count (12.25 per stroke) with uniform intervals. The soft MP suppressed the volume decay, yielding 6.8-times higher acoustic efficiency compared to the rigid MP. These results indicate that the proposed tricks with the combination of elasticity and optimization can make robots play excellent drum rolls.
I. INTRODUCTION
High-frequency drum rolls combine active strokes with passive bounces, but robotic reproduction is limited by hardware constraints and nonlinear soft-body behavior. This paper addresses these issues with two optimized active-motion tricks for soft robotic drumming.
- Drum rolls combine active strokes and passive bounces to produce continuous, white-noise-like sound above 30 Hz.
- Robotic drum-roll reproduction is difficult because friction, backlash, sensing latency, and related hardware limitations hinder high-speed operation.
- Elasticity supports passive rebounds, but optimal stiffness and control are difficult to analyze because of nonlinear and uncertain physical factors.
- The paper identifies limited waveform shaping from passive elasticity and low sample efficiency in real-world optimization as open issues.
- The Tap-Pull trick prolongs rebounds, while Micro-Pulse compensates amplitude decay; Bayesian optimization tunes their parameters from sound-based objectives.
- Optimized tricks increased bounce count to over 12 and improved acoustic efficiency by 6.8 times in the dual-arm robot.
II. SYSTEM SETUP
The system uses a dual-arm robot with interchangeable soft and rigid end-effectors, while microphone recordings provide the acoustic data for evaluating drum rolls.
- Each robot arm has three joints, and joint-angle commands are updated at 200 Hz for repetitive collision tasks.
- Soft PLA end-effectors use an elastic rubber ball to generate passive bounces, whereas reinforced rigid arms suppress tip rebound.
- A condenser microphone records single-channel acoustic signals at 44.1 kHz with 1024-sample blocks.
- The recorded waveform is band-passed to 1–100 Hz, transformed into a smoothed envelope, and analyzed for impact peaks.
III. ACTIVE MOTION TRICKS
The paper defines drum-roll objectives around bounce count, interval uniformity, amplitude uniformity, and efficiency, then augments alternating step-function strokes with active motion tricks.
- A high-quality drum roll favors even bounce intervals, even amplitudes, and rapid white-noise-like impacts.
- The two target behaviors are maximal uniform bounces and uniform amplitudes with uniform intervals under a constant bounce count.
- The basic trajectory alternates the two arms vertically with period T, producing one impact in every period.
- A larger upper-to-lower position difference qualitatively produces louder sound, while a smaller lower position can increase impact but affect resonance and later bounces.
- The Tap-Pull trick adds horizontal retraction after a downward tap to inject energy into passive rebounds and sustain continuous sound.
C. Tap-Pull trick
Tap-Pull augments a normal downward strike with timed horizontal retraction, using displacement and timing parameters to inject energy into subsequent rebounds.
- Tap-Pull has two phases: a vertical downward Tap followed by a horizontal Pull toward the robot.
- The Pull begins after waiting time tdown during the downstroke and ends with total displacement dxpull.
- The pulling motion injects kinetic energy needed for bouncing, allowing the drum roll to continue without interruption.
- The horizontal command moves linearly by dxpull after waiting tdown, with pull duration tpull = T/2 − tdown.
D. Micro-Pulse trick
The Micro-Pulse trick injects fine oscillations during each downstroke to excite elastic-arm vibrations and maintain drum-roll loudness. Its amplitude and frequency parameters determine the pulse pattern without changing the overall stroke period.
- D. Micro-Pulse trick: MP inserts small-amplitude oscillations during each downstroke to excite vibrations in the elastic end-effector.The trick targets more uniform sound amplitude by actively stimulating the arm’s elastic dynamics.
- D. Micro-Pulse trick: The MP command excites elastic dynamics without changing the overall stroke period.The motion is intended to increase striking speed and enhance impact sound without decay.
- D. Micro-Pulse trick: Micro pulses are inserted Nµ times, with Nµ controlling the frequency of the resulting micro-vibrations.The downstroke is divided into 2(Nµ−1)+1 segments to schedule the pulses.
- D. Micro-Pulse trick: The micro-pulse amplitude Aµ is greater than the baseline pulse amplitude AL.The command alternates between the baseline level and the higher-amplitude micro pulse within each segment.
IV. OPTIMIZATION OF TRICKS
The study tunes TP and MP parameters through data-driven Bayesian optimization because soft-robot dynamics are nonlinear and difficult to analyze directly. Objectives are defined from acoustic outcomes, including bounce count, interval uniformity, and sound-volume decay.
- IV. OPTIMIZATION OF TRICKS: TP and MP parameters are optimized independently because nonlinear soft-robot dynamics make their effects difficult to analyze.The study automates parameter search in a data-driven manner.
- IV. OPTIMIZATION OF TRICKS: Bayesian optimization provides sample-efficient black-box tuning for the feedforward tricks.The implementation uses a Tree-Structured Parzen Estimator sampler in Optuna.
- IV. OPTIMIZATION OF TRICKS: The TP objective encourages more bounces while improving the uniformity of bounce intervals.It combines bounce count N with interval-uniformity metric γint using weighting coefficient λTP = 5.
- IV. OPTIMIZATION OF TRICKS: Bounce count N is measured as the number of acoustic peaks detected during one stroke.The interval-uniformity metric γint is defined from inter-bounce intervals and is smaller when those intervals are more uniform.
C. Optimization for MP trick
The MP trick is optimized to suppress post-impact amplitude decay while preserving bounce timing, target bounce count, and acoustic efficiency. Bayesian optimization minimizes a weighted objective combining these criteria and motor power.
- Objective design: The MP objective combines amplitude decay, interval uniformity, target bounce count, and motor power through weighted penalties.The weights are λγ = 0.5, λN = 0.2, and λP = 8.0, with fixed upper and lower amplitudes of 0.04 m and 0 m.
- Objective design: The desired MP behavior suppresses amplitude decay after the first strike while maintaining uniform bounce intervals.The bounce count is controlled to a target value so efficiency and tempo can be compared fairly.
- Objective design: Amplitude decay is measured from the impact amplitudes A_k, with the first-impact amplitude A_1 serving as the reference.The decay ratio should be zero to stabilize sound while injecting the minimum necessary energy.
- Efficiency measure: Acoustic efficiency is defined as the ratio of acoustic power to motor power, but uncertain acoustic constants make its numerical estimation unreliable.The study therefore uses an alternative efficiency measure for relative evaluation and adds only motor-power minimization to the objective.
A. Setup
The evaluation compares soft and rigid arms with and without TP or MP tricks, optimizing each hardware–trick condition independently. The soft TP condition produced the highest bounce count and most uniform intervals, while TP also enabled continuous sound between strokes.
- A. Setup: Four hardware–trick conditions compare soft and rigid arms with and without each trick, using independent Bayesian optimization.TP settings were optimized for 200 iterations and MP settings for 100 iterations.
- A. Setup: Four strokes were executed by the left and right arms during trials to verify that the drum roll continued continuously.The trials used both arms rather than evaluating only a single isolated stroke.
- B. TP trick: The soft arm with TP achieved the highest bounce count, reaching 12.25, and also had the most uniform bounce intervals.Introducing TP increased bounce count for both rigid and soft arms.
- B. TP trick: TP produced less stable bouncing on the rigid arm because its tip was not flexible enough for natural rebound acoustics.The rigid-arm result was described as more like dragging than natural bouncing.
- B. TP trick: Without TP, bouncing on the soft arm ended before the arm swung upward, creating silent gaps between strokes.TP enabled continuous sound between strokes, which is desirable for drum rolls.
C. MP trick
The MP trick uses small oscillations to sustain post-impact sound while preserving the specified bounce pattern. Optimized soft MP substantially reduced volume decay and delivered higher acoustic efficiency than rigid MP, despite increased motor power.
- The optimized MP trick kept both rigid and soft arms at the specified five bounces while improving inter-bounce uniformity.
- Soft MP significantly reduced sound-volume decay, whereas rigid MP could not reduce it sufficiently.
- About 683% improvement in acoustic efficiency favored soft MP over rigid MP.Soft MP required more motor power because the motors injected dissipated energy.
- The highest acoustic efficiency occurred without MP because of excessive bounce count, making optimized soft MP the best-organized and most efficient roll.
VI. DISCUSSION
The discussion argues that soft embodiment makes parameter tuning highly sensitive, motivating Bayesian optimization to exploit its potential. It also identifies limits in experimental scope and perceptual evaluation, while proposing broader multi-objective optimization.
- Soft-arm performance was highly sensitive to TP and MP parameters, unlike rigid-arm performance, making manual optimization difficult.
- Bayesian optimization is important for exploiting soft embodiment because it estimates objective values across the parameter space.
- Experiments used a fixed drum configuration and limited tempo range, so parameter-sensitivity generality across physical setups remains unverified.
- The quantitative objective did not explicitly evaluate perceptual drum-roll quality, leaving its relationship to human judgments for further investigation.
- Future work could optimize sound amplitude, interval uniformity, energy efficiency, and bounce count jointly for greater expressivity.