SpikePingpong: Spike Vision-based Fast-Slow Pingpong Robot System
Hao Wang, Chengkai Hou, Xianglong Li, Yankai Fu, Chenxuan Li, Ning Chen, Gaole Dai, Jiaming Liu, Tiejun Huang, Shanghang Zhang

TL;DR
SpikePingpong is a novel spike vision-based robotic system that combines fast perception and precise control to excel in high-speed table tennis tasks, demonstrating high success rates in target hitting accuracy.
Contribution
It introduces a Fast-Slow architecture integrating spike-based vision and imitation learning for high-precision, time-critical robotic table tennis control.
Findings
Achieves 92% success rate in 30cm accuracy zones
Attains 70% success in 20cm precision targeting
Demonstrates effectiveness of spike vision in dynamic tasks
Abstract
Learning to control high-speed objects in dynamic environments represents a fundamental challenge in robotics. Table tennis serves as an ideal testbed for advancing robotic capabilities in dynamic environments. This task presents two fundamental challenges: it requires a high-precision vision system capable of accurately predicting ball trajectories under complex dynamics, and it necessitates intelligent control strategies to ensure precise ball striking to target regions. High-speed object manipulation typically demands advanced visual perception hardware capable of capturing rapid motion with exceptional temporal resolution. Drawing inspiration from Kahneman's dual-system theory, where fast intuitive processing complements slower deliberate reasoning, there exists an opportunity to develop more robust perception architectures that can handle high-speed dynamics while maintaining…
Peer Reviews
Decision·ICLR 2026 Poster
+ Well-structured and intuitive architecture. The modular pipeline from perception to control is clear and logically connected. + Detailed modular implementation. The paper provides comprehensive implementation details for each subsystem, enabling reproducibility and practical insights for future robotic applications. + Good real-world results. Experimental evaluation on the physical ABB IRB-120 setup demonstrates reliable tracking, accurate returns, and stable rally performance, confirming re
The main weakness of this paper lies in its limited adaptation and generalization capability. Since the Stage-3 action generation (IMPACT) relies purely on imitation learning, the system lacks the ability to adapt to unseen or out-of-distribution ball trajectories, such as those with different spins, velocities, or bounce patterns. The evaluation also appears to be confined to in-domain scenarios, and it remains unclear whether the tests include unseen human/robot launch opponents or novel shot
- The paper is overall well-written and easy to follow. - It is impressive to see the demo included in the supplementary submission, which demonstrates the robustness of the system. - The experiments are comprehensive to demonstrate the importance of the design choices made in this work as well as the overall robustness of the system. - As a system-oriented paper, the design choices and their details presented in the paper can be useful references for future works in similar direction.
- No major weaknesses. - Typo in Figure 2: “inversive kinamatics”
- The Fast-Slow system architecture is well-motivated and technically sound, combining physics-based modeling with learned corrections. - The spike camera integration for capturing millisecond-level ball-paddle contact is innovative and well-executed. - The experimental validation includes solid real-world deployment and comprehensive comparisons and ablations.
- The task focuses solely on table tennis. There are many task-specific design choices with insufficient analysis of what methodology generalizes beyond table tennis (and applies to more general dynamic manipulation tasks). - There lacks a in-depth failure analysis. Table 8 shows distributions but doesn't explain root causes of 79.1% near-miss failures or how failures vary by ball speed, spin, trajectory. - Minor: Wrong citation in Table 3: "Diffusion Policy (Zhao et al., 2023)" should be "(Chi
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsRobot Manipulation and Learning · Soft Robotics and Applications · Reinforcement Learning in Robotics
