TaCo: A Benchmark for Lossless and Lossy Codecs of Heterogeneous Tactile Data

TL;DR

TaCo introduces the first comprehensive benchmark for tactile data codecs, addressing the challenge of compressing heterogeneous tactile signals under real-time constraints. It evaluates 30 codecs across five tactile datasets and pairs traditional off-the-shelf methods with neural and data-driven approaches, including the newly trained TaCo-LL (lossless) and TaCo-L (lossy) codecs. Across four tasks—lossless storage, human visualization, semantic classification, and dexterous grasping—TaCo-LL and TaCo-L demonstrate superior performance and reveal meaningful trade-offs between compression efficiency and downstream task fidelity. The framework enables cross-domain insights and sets a practical foundation for scalable tactile perception and transmission in robotics.

Abstract

Tactile sensing is crucial for embodied intelligence, providing fine-grained perception and control in complex environments. However, efficient tactile data compression, which is essential for real-time robotic applications under strict bandwidth constraints, remains underexplored. The inherent heterogeneity and spatiotemporal complexity of tactile data further complicate this challenge. To bridge this gap, we introduce TaCo, the first comprehensive benchmark for Tactile data Codecs. TaCo evaluates 30 compression methods, including off-the-shelf compression algorithms and neural codecs, across five diverse datasets from various sensor types. We systematically assess both lossless and lossy compression schemes on four key tasks: lossless storage, human visualization, material and object classification, and dexterous robotic grasping. Notably, we pioneer the development of data-driven codecs explicitly trained on tactile data, TaCo-LL (lossless) and TaCo-L (lossy). Results have validated the superior performance of our TaCo-LL and TaCo-L. This benchmark provides a foundational framework for understanding the critical trade-offs between compression efficiency and task performance, paving the way for future advances in tactile perception.

Figures (19)

• Figure 1: The motivation of our TaCo benchmark, established through an extensive evaluation on tactile codecs across multiple dimensions. First, we assess 30 off-the-shelf and neural codecs on 5 heterogeneous tactile datasets with more than 250K frames. Second, we introduce purely-trained TaCo-LL and TaCo-L codecs to explore the data-driven approaches in the field of lossless and lossy tactile data compression. Finally, we evaluate the coding performance on 4 distinct task types designed to serve for human, machine, and robotics.
• Figure 2: Diagram of data-driven compression methods and our proposed TaCo-L and TaCo-LL.
• Figure 3: Detailed implementations of our proposed TaCo-L and TaCo-LL.
• Figure 4: Rate-distortion curves on TouchandGo dataset, when applying intra-frame compression methods.
• Figure 5: Visualization of tactile data characteristics across different datasets, sensors, and object types. Each subfigure displays the raw tactile image, its frequency spectrum, and the corresponding entropy map.