Robotics Lab

Operational Space Control

A key hardware capability developed at our lab is joint torque control, which enables precise yet compliant motion at the robot’s hands, and improves safety in the event of unexpected collisions. Joint torque control also greatly simplifies controller design, especially for complex robots. This control methodology, named operational space control has laid the foundation for robot capabilities such as dexterous collision avoidance and posture control. In an industrial application, we programmed a robot arm to assemble various parts of an airplane’s wingbox in a highly confined space through the use of redundant motion. 

Constraint-Consistent Whole Body Control

Autonomous assistants such as humanoid robots require reactive realtime controllers which can perform interactive tasks at their hands while satisfying constraints such as balance, collision avoidance and joint limits. We extended operational space control to a control methodology that enables realtime satisfaction of these constraints while allowing easy task programming at the robot's hands. This framework is called the constraint-consistent whole body control, and was successfully deployed on as ASIMO robot from Honda. The study of robot programming and behavior inspires an appreciation for the complex working of humans. Similarly, robots can benefit much from the study of human motion control. Our work in musculoskeletal modeling has demonstrated that humans optimize their posture to perform tasks such as pushing a heavy object to maximize energy efficiency. This has led to better energy efficiency in robot whole body control. 

Compliant Motion Primitive

People are also very skilled in object manipulation. We created a mathematical language of robot skill description called the compliant motion primitive. Complex skills such as plugging a power cord into a socket can be decomposed into several simpler primitives. In challenging unstructured environments where the robot must handle many new objects and workspace configurations every day, we believe that skill based programming is the only scalable approach to autonomy. Currently, we are working on a sponsored project to automate warehouse operations such as item storage and retrieval from boxes through a library of compliant motion primitives and skill based programming. 

Two Hands

The lack of fixtures in human and natural environments necessitates the use of two hands. Two hands mounted on two separate arms brings new coordination challenges that must be addressed for safe behavior. We developed the augmented object model to facilitate object level control independent of the number of arms grasping the object. Together with a whole body controller, this model allows the seamless control of two hand robotic systems that perform complex real world tasks such as lifting a large box by pushing against the sides. Through an ongoing project, we hope to establish many useful two hand behaviors for object manipulation in our homes.

Ocean One

A unique story that emerged from the different research directions we pursued over the last four decades is that of our underwater humanoid robot, Ocean One. Designed and developed at the Robotics Lab in collaboration with other labs and companies, Ocean One brings a novel capability of fine manipulation at depths up to 1000m for mechanical, biological and archaeological applications. In 2016, the robot recovered a French national treasure from the Lune ship wrecked in 1664 to the floor of the Mediterranean Sea at 91m depth. Ocean One is guided by an expert located on the boat through an intuitive haptic interface. We envision a future where robots like Ocean One can aid in the safe exploration and exploitation of remote spaces for human progress.