Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Learning Variable Compliance Control From a Few Demonstrations for Bimanual Robot with Haptic Feedback Teleoperation System (2406.14990v2)

Published 21 Jun 2024 in cs.RO, cs.AI, and cs.LG

Abstract: Automating dexterous, contact-rich manipulation tasks using rigid robots is a significant challenge in robotics. Rigid robots, defined by their actuation through position commands, face issues of excessive contact forces due to their inability to adapt to contact with the environment, potentially causing damage. While compliance control schemes have been introduced to mitigate these issues by controlling forces via external sensors, they are hampered by the need for fine-tuning task-specific controller parameters. Learning from Demonstrations (LfD) offers an intuitive alternative, allowing robots to learn manipulations through observed actions. In this work, we introduce a novel system to enhance the teaching of dexterous, contact-rich manipulations to rigid robots. Our system is twofold: firstly, it incorporates a teleoperation interface utilizing Virtual Reality (VR) controllers, designed to provide an intuitive and cost-effective method for task demonstration with haptic feedback. Secondly, we present Comp-ACT (Compliance Control via Action Chunking with Transformers), a method that leverages the demonstrations to learn variable compliance control from a few demonstrations. Our methods have been validated across various complex contact-rich manipulation tasks using single-arm and bimanual robot setups in simulated and real-world environments, demonstrating the effectiveness of our system in teaching robots dexterous manipulations with enhanced adaptability and safety. Code available at: https://github.com/omron-sinicx/CompACT

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Tatsuya Kamijo (3 papers)
  2. Cristian C. Beltran-Hernandez (12 papers)
  3. Masashi Hamaya (15 papers)
Citations (3)

Summary

Overview of "Learning Variable Compliance Control From a Few Demonstrations for Bimanual Robot with Haptic Feedback Teleoperation System"

This paper presents an innovative framework to facilitate contact-rich manipulation tasks for rigid robots through a system leveraging Virtual Reality (VR) controllers and advanced control methods. The primary aim is to teach robots to perform complex manipulation tasks by drawing on only a limited number of demonstrations, offering improvements in safety, adaptability, and intuitiveness through a Learning from Demonstrations (LfD) approach. Critically, the authors address the challenge of rigidity in traditional robotic systems, which often result in excessive contact forces, by incorporating a Variable Compliance Control strategy within a framework termed Comp-ACT (Compliance Control via Action Chunking with Transformers).

Key Contributions

The authors propose two primary contributions:

  1. A VR-Inspired Teleoperation System: This system includes a cost-effective interface using VR controllers to intuitively demonstrate tasks while integrating haptic feedback via vibrations. This hardware setup allows operators to teach robots by leveraging natural human hand movements, addressing both user-friendliness and cost concerns.
  2. Comp-ACT for Variable Compliance Control: Comp-ACT enhances robot learning by integrating a prediction model using Transformers. This model predicts future actions in sequences rather than single steps, thus minimizing compounding errors and allowing the robots to learn task-specific compliance characteristics effectively. Furthermore, this work is distinguished from past approaches by focusing on task-space trajectory learning and including stiffness parameter predictions alongside motion trajectories without requiring direct torque control.

Experimental Validation

The proposed method was subjected to rigorous testing in both simulated and real-world settings. The experiments addressed various dexterous tasks including bimanual wiping, picking and inserting, wiping and drawing on surfaces, and peg-in-hole tasks with complex geometries:

  • Simulation Results: Compared to basic ACT methods, Comp-ACT resulted in a substantially lower mean contact force when performing wiping tasks. Demonstrations highlighted the method's proficiency in mitigating excessive force application by maintaining variable compliance during contact-rich interactions.
  • Real-World Experiments: Success in five manipulation tasks showcases the proficient generalization of Comp-ACT to practical scenarios. Results such as a 100% success rate in tasks like single-arm drawing indicate robust performance. The integration of force/torque data as observational input further strengthens the model’s adaptability, especially pronounced in bimanual peg-in-hole tasks.

Implications and Future Directions

The introduction of a VR-mounted teleoperation and the Comp-ACT model has several implications for the field of robotics. The potential for reducing demonstration costs while increasing robot adaptability presents significant practical improvements. Moreover, independent manipulation of compliance benefits worker safety and task versatility, pivotal for environments requiring human-robot interaction or those involving delicate maneuvers.

Future research will need to extend this framework's adaptability to multiple settings by enabling the system to autonomously adjust compliance based on real-time feedback. Also noteworthy is the potential to expand this work towards broader multi-tasking capabilities or more generalized manipulation scenarios, potentially incorporating broader sensor inputs or integrating with other advanced learning methodologies. Exploring real-world application limits and continuing to refine the compliance interface should yield further insights into the robustness of this approach.

This paper thus sets a critical foundation for advancing dexterous robotic manipulation, leveraging recent advances in LfD and human-friendly interfaces to mitigate classical challenges in robotic compliance control.

Youtube Logo Streamline Icon: https://streamlinehq.com