Evidence base
Three converging arguments from literature support the role of tactile sensing in contact-critical manipulation.
The case for tactile sensing rests on documented failure modes. Research in robotics and vision systems has identified specific task classes where vision-based approaches hit structural limits. Other work has shown that tactile information provides calibration that vision cannot access. And combined tactile-vision systems have demonstrated improved performance over vision alone.
Zheng et al. 2025, EgoScale, arXiv:2602.16710
Evaluated 59 vision-language models on dexterous manipulation tasks. Card manipulation and force closure tasks showed the highest failure rates. These are precisely the task classes where the contact interface is occluded during execution.
Forces for Free, Science Robotics 2025, adq5046
Tactile sensing provides calibration signals that vision cannot access. The contact interface contains force distribution information essential for stable manipulation that cannot be inferred from visual observation alone.
NeuralFeels (Science Robotics 2024); 3D-ViTac (arXiv:2410.24091)
Systems combining tactile and vision inputs outperform vision-only approaches on manipulation tasks. The modalities are complementary: vision provides context and pose estimation; tactile provides contact state and force information.
Sensor characterization targeting benchmark-grade measurements.
OVR is currently in the characterization phase. The focus is on building toward benchmark-grade data across the three application domains: occluded grasping, in-hand reorientation, and contact-critical force tasks.
Data collection methodology is being defined. The goal is measurements that can support claims about sensor performance and control loop effectiveness in these task classes.
Open to benchmark co-design with labs working on dexterous manipulation.
If your research group is working on dexterous manipulation benchmarks, tactile-vision integration, or contact-rich control problems, we are interested in discussing co-design of evaluation protocols.
The thesis that tactile measurement improves manipulation reliability in contact-critical tasks is stronger when tested against shared benchmarks. We are open to collaboration on methodology definition and joint evidence-building work.
Current state
Characterization phase
Active work on sensor characterization and data collection methodology. Not yet collecting benchmark data. Research collaboration inquiries welcome.