Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
120 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
46 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Adaptive Whole-body Robotic Tool-use Learning on Low-rigidity Plastic-made Humanoids Using Vision and Tactile Sensors (2405.04826v1)

Published 8 May 2024 in cs.RO

Abstract: Various robots have been developed so far; however, we face challenges in modeling the low-rigidity bodies of some robots. In particular, the deflection of the body changes during tool-use due to object grasping, resulting in significant shifts in the tool-tip position and the body's center of gravity. Moreover, this deflection varies depending on the weight and length of the tool, making these models exceptionally complex. However, there is currently no control or learning method that takes all of these effects into account. In this study, we propose a method for constructing a neural network that describes the mutual relationship among joint angle, visual information, and tactile information from the feet. We aim to train this network using the actual robot data and utilize it for tool-tip control. Additionally, we employ Parametric Bias to capture changes in this mutual relationship caused by variations in the weight and length of tools, enabling us to understand the characteristics of the grasped tool from the current sensor information. We apply this approach to the whole-body tool-use on KXR, a low-rigidity plastic-made humanoid robot, to validate its effectiveness.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (19)
  1. K. Hirai, M. Hirose, Y. Haikawa, and T. Takenaka, “The Development of Honda Humanoid Robot,” in Proceedings of the 1998 IEEE International Conference on Robotics and Automation, 1998, pp. 1321–1326.
  2. K. Kaneko, F. Kanehiro, S. Kajita, H. Hirukawa, T. Kawasaki, M. Hirata, K. Akachi, and T. Isozumi, “Humanoid robot HRP-2,” in Proceedings of the 2004 IEEE International Conference on Robotics and Automation, 2004, pp. 1083–1090.
  3. P. Allgeuer, H. Farazi, M. Schreiber, and S. Behnke, “Child-sized 3D printed igus humanoid open platform,” in Proceedings of the 2015 IEEE-RAS International Conference on Humanoid Robots, 2015, pp. 33–40.
  4. T. Makabe, T. Anzai, Y. Kakiuchi, K. Okada, and M. Inaba, “Development of Amphibious Humanoid for Behavior Acquisition on Land and Underwater,” in Proceedings of the 2020 IEEE-RAS International Conference on Humanoid Robots, 2020, pp. 104–111.
  5. C. Lee, M. Kim, Y. J. Kim, N. Hong, S. Ryu, H. J. Kim, and S. Kim, “Soft robot review,” International Journal of Control, Automation and Systems, vol. 15, no. 1, pp. 3–15, 2017.
  6. C. T. Kiang, A. Spowage, and C. K. Yoong, “Review of Control and Sensor System of Flexible Manipulator,” Journal of Intelligent & Robotic Systems, vol. 77, no. 1, pp. 187–213, 2015.
  7. C. Sun, W. He, and J. Hong, “Neural Network Control of a Flexible Robotic Manipulator Using the Lumped Spring-Mass Model,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 47, no. 8, pp. 1863–1874, 2017.
  8. J. Y. Loo, Z. Y. Ding, V. M. Baskaran, S. G. Nurzaman, and C. P. Tan, “Robust multimodal indirect sensing for soft robots via neural network-aided filter-based estimation,” Soft Robotics, vol. 9, no. 3, pp. 591–612, 2022.
  9. G. Runge, M. Wiese, and A. Raatz, “FEM-based training of artificial neural networks for modular soft robots,” in Proceedings of the 2017 IEEE International Conference on Robotics and Biomimetics, 2017, pp. 385–392.
  10. T. G. Thuruthel, B. Shih, C. Laschi, and M. T. Tolley, “Soft robot perception using embedded soft sensors and recurrent neural networks,” Science Robotics, vol. 4, no. 26, p. eaav1488, 2019.
  11. J. Tani, “Self-organization of behavioral primitives as multiple attractor dynamics: a robot experiment,” in Proceedings of the 2002 International Joint Conference on Neural Networks, 2002, pp. 489–494.
  12. K. P. Tee, J. Li, L. T. P. Chen, K. W. Wan, and G. Ganesh, “Towards Emergence of Tool Use in Robots: Automatic Tool Recognition and Use Without Prior Tool Learning,” in Proceedings of the 2018 IEEE International Conference on Robotics and Automation, 2018, pp. 6439–6446.
  13. A. Xie, F. Ebert, S. Levine, and C. Finn, “Improvisation through Physical Understanding: Using Novel Objects as Tools with Visual Foresight,” arXiv preprint arXiv:1904.05538, 2019.
  14. C. Nabeshima, Y. Kuniyoshi, and M. Lungarella, “Adaptive body schema for robotic tool-use,” Advanced Robotics, vol. 20, no. 10, pp. 1105–1126, 2006.
  15. Y. Wu, K. Takahashi, H. Yamada, K. KIM, S. Murata, S. Sugano, and T. Ogata, “Dynamic Motion Generation by Flexible-Joint Robot based on Deep Learning using Images,” in Proceeding of the 8th Joint IEEE International Conference on Development and Learning on Epigenetic Robotics, 2018.
  16. M. Eppe, P. D. H. Nguyen, and S. Wermter, “From Semantics to Execution: Integrating Action Planning With Reinforcement Learning for Robotic Causal Problem-Solving,” Frontiers in Robotics and AI, vol. 6, p. 123, 2019.
  17. K. Fang, Y. Zhu, A. Garg, A. Kurenkov, V. Mehta, L. Fei-Fei, and S. Savarese, “Learning task-oriented grasping for tool manipulation from simulated self-supervision,” The International Journal of Robotics Research, vol. 39, no. 2-3, pp. 202–216, 2020.
  18. K. Kawaharazuka, K. Okada, and M. Inaba, “Adaptive Robotic Tool-Tip Control Learning Considering Online Changes in Grasping State,” IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 5992–5999, 2021.
  19. D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” in Proceedings of the 3rd International Conference on Learning Representations, 2015, pp. 1–15.
Citations (1)
View on Semantic Scholar

Summary

  • The paper details a neural network that uses Parametric Bias to adjust for tool weight and length variations in soft humanoid robots.
  • It employs online sensor fusion and adaptive control to manage dynamic shifts in tool-tip positioning and center of gravity.
  • Experimental results on the KXR robot demonstrate the method’s ability to adapt to diverse tools without manual recalibration.

Exploring Whole-Body Tool-Use Learning for Low-Rigidity Robots

Introduction to Low-Rigidity Robot Challenges

Robots come in various shapes and sizes, each with their own unique capabilities and challenges. While high-rigidity robots excel in industrial settings due to their precision and strength, low-rigidity robots, often made from materials like plastic or rubber, face distinct challenges, particularly when it comes to tool use. These robots can experience significant deflections that alter their tool-tip positions and their center of gravity when interacting with objects, complicating tasks that involve manipulation and precise positioning.

The Study’s Approach

The paper introduces a novel neural network method that integrates joint angle, visual, and tactile data to manage the complex dynamics of low-rigidity robots during tool-use. This method leverages what's termed as Parametric Bias (PB) to account for changes in tool weight and length, aiming to improve the robot's interaction with its environment by predicting and adjusting to these variances in real-time.

Key Components and Innovations

The system, dubbed Whole-body Tool-use Network with Parametric Bias (WTNPB), is structured to dynamically adjust to the tool's physical characteristics, which are learned through the robot's sensory inputs. Key components include:

  • Data Collection and Network Training: The network is trained using data from various tool states, with a special emphasis on capturing changes caused by different tool weights and lengths.
  • Parametric Bias: This is a mechanism that captures variations in tool weight and length, thus aiding the robot in adjusting its movements according to the changes in the tool's properties.
  • Online Update Capability: The WTNPB supports online updates, allowing the robot to adjust to new tools in real-time by updating the PB based on current sensor data.
  • Control Strategy: The robot can control the tool-tip position by optimizing the network outputs with respect to the desired objectives, such as maintaining a specific position while accounting for shifts in the center of gravity.

Experimental Setup and Results

The researchers used a low-rigidity humanoid robot named KXR, equipped with sensors and markers to collect necessary data. The network was tested in various scenarios—including simulations and actual robot experiments—to demonstrate its effectiveness in adapting to different tools and conditions. The results from these experiments showed that the WTNPB could successfully adjust its internal model to effectively handle tools of various weights and lengths without manual recalibration.

Theoretical and Practical Implications

Practically, this research paves the way for more adaptive and flexible robotic systems in environments where interaction with various objects is necessary. Theoretically, it introduces an integrated approach to handling the variability in physical interactions caused by robot flexibility, suggesting that complex, real-world tasks could be managed more effectively with advanced neural network architectures.

Future Directions

Looking ahead, there are several avenues for expansion and improvement:

  1. Improving the Automatic Learning of Mask Variables: Current systems require predefined mask variables. Automating this process could lead to more adaptive systems that can learn optimal configurations from their experiences.
  2. Scaling to Larger Robots: While this paper focused on a small humanoid robot, applying these techniques to larger, more complex systems could address broader applications, including industrial automation and more sophisticated service robots.
  3. Enhancing Stability and Safety: As robots become more autonomous and capable of handling a variety of tools and tasks, ensuring their operational stability and safety, especially in unpredictable environments, remains a priority.

This paper contributes significantly to our understanding of robotic control and interaction in the context of low-rigidity systems and sets a foundation for future work looking to expand these principles to more complex and varied robot designs.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
Youtube Logo Streamline Icon: https://streamlinehq.com

YouTube