The field of robotics is witnessing significant advancements in active tactile perception and multimodal learning, with a growing emphasis on integrating tactile sensing with vision and language models. Researchers are exploring novel frameworks that leverage reinforcement learning, transformer-based architectures, and visuotactile fusion to improve robot manipulation and perception capabilities. Notably, the development of task-agnostic active perception frameworks and language-tactile pretraining models is enabling more effective and generalizable solutions for contact-rich tasks. These innovative approaches are demonstrating superior performance in various tasks, including object classification, shape reconstruction, and human activity recognition. Furthermore, the integration of multimodal sensing and learning is enhancing the robustness and adaptability of robotic systems in complex environments. Some particularly noteworthy papers in this area include:

• A novel framework that introduces a task-agnostic active perception approach, achieving high accuracies in haptic digit recognition and tactile pose estimation tasks.
• A visuotactile fusion framework that enhances robotic manipulation under visual constraints, outperforming baselines in contact-rich tasks such as surface wiping and peg insertion.
• A language-tactile pretraining model that aligns tactile 3D point clouds with natural language, enabling contact-state-aware tactile language understanding for manipulation tasks.
• A multimodal framework that combines tactile and motion data for human activity recognition, consistently outperforming single-modality methods.
• A vision-tactile-language-action model that enables robust policy generation in contact-intensive scenarios, achieving over 90% success rates in unseen peg shapes.