Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards a Robust Soft Baby Robot With Rich Interaction Ability for Advanced Machine Learning Algorithms

Mohannad Alhakami, Dylan R. Ashley, Joel Dunham, Yanning Dai, Francesco Faccio, Eric Feron, Jürgen Schmidhuber

TL;DR

Problem: modern reinforcement learning in robotics requires platforms capable of long autonomous exploration with rich sensory input while tolerating sensor and actuation failures. Approach: a starfish-inspired, hybrid soft-hard robotic limb with camera-only sensing, high sensor redundancy, and replaceable failure points is built and validated in hardware; PPO and AC are tested under simulated camera failures to demonstrate robust learning. Contributions: a modular hardware design and fabrication workflow, a tendon-driven soft outer limb, camera-based sensing architecture, and empirical evidence of resilient RL under degraded sensing, complemented by open-source CAD, Gazebo models, and code. Significance: this work advances a practical path toward curiosity-driven learning and generally intelligent robots by decoupling learning from fragile hardware constraints.

Abstract

Advanced machine learning algorithms require platforms that are extremely robust and equipped with rich sensory feedback to handle extensive trial-and-error learning without relying on strong inductive biases. Traditional robotic designs, while well-suited for their specific use cases, are often fragile when used with these algorithms. To address this gap -- and inspired by the vision of enabling curiosity-driven baby robots -- we present a novel robotic limb designed from scratch. Our design has a hybrid soft-hard structure, high redundancy with rich non-contact sensors (exclusively cameras), and easily replaceable failure points. Proof-of-concept experiments using two contemporary reinforcement learning algorithms on a physical prototype demonstrate that our design is able to succeed in a simple target-finding task even under simulated sensor failures, all with minimal human oversight during extended learning periods. We believe this design represents a concrete step toward more tailored robotic designs for achieving general-purpose, generally intelligent robots.

Towards a Robust Soft Baby Robot With Rich Interaction Ability for Advanced Machine Learning Algorithms

TL;DR

Problem: modern reinforcement learning in robotics requires platforms capable of long autonomous exploration with rich sensory input while tolerating sensor and actuation failures. Approach: a starfish-inspired, hybrid soft-hard robotic limb with camera-only sensing, high sensor redundancy, and replaceable failure points is built and validated in hardware; PPO and AC are tested under simulated camera failures to demonstrate robust learning. Contributions: a modular hardware design and fabrication workflow, a tendon-driven soft outer limb, camera-based sensing architecture, and empirical evidence of resilient RL under degraded sensing, complemented by open-source CAD, Gazebo models, and code. Significance: this work advances a practical path toward curiosity-driven learning and generally intelligent robots by decoupling learning from fragile hardware constraints.

Abstract

Advanced machine learning algorithms require platforms that are extremely robust and equipped with rich sensory feedback to handle extensive trial-and-error learning without relying on strong inductive biases. Traditional robotic designs, while well-suited for their specific use cases, are often fragile when used with these algorithms. To address this gap -- and inspired by the vision of enabling curiosity-driven baby robots -- we present a novel robotic limb designed from scratch. Our design has a hybrid soft-hard structure, high redundancy with rich non-contact sensors (exclusively cameras), and easily replaceable failure points. Proof-of-concept experiments using two contemporary reinforcement learning algorithms on a physical prototype demonstrate that our design is able to succeed in a simple target-finding task even under simulated sensor failures, all with minimal human oversight during extended learning periods. We believe this design represents a concrete step toward more tailored robotic designs for achieving general-purpose, generally intelligent robots.
Paper Structure (17 sections, 8 figures, 1 table)

This paper contains 17 sections, 8 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. BACKGROUND AND RELATED WORK
  3. ROBOTIC LIMB DESIGN
  4. Methodology
  5. Materials selection
  6. Mechanical Design
  7. Cameras and Sensors
  8. Electronics Design
  9. Wiring and Tendon Challenges
  10. Mechanical Testing
  11. MACHINE LEARNING EXPERIMENT
  12. Methodology
  13. Results
  14. Discussion
  15. CONCLUSIONS AND FUTURE WORK
  16. ...and 2 more sections

Figures (8)

  • Figure 1: The design of the limb we construct is loosely inspired by the shape of a starfish in that we envision a multi-limbed headless creature with the bulk of the sensors being contained in the limbs themselves.
  • Figure 2: Exploded view of the Robot with colour-coding based on function---segments (red), brackets (black), camera enclosures (green), cable management (blue), and attachments(gray).
  • Figure 3: The effect on the field of view for the four front-facing cameras when the limb is extended and retracted.
  • Figure 4: Wiring Diagram including the Arty A7 development board for networking.
  • Figure 5: The custom PCB for combining ESP32-Cam(s), sensors, and power/networking wiring management.
  • ...and 3 more figures