Differentiable Simulation Methods for Robotic Agent Design
Host
Wojciech Matusik
MIT EECS/CSAIL
Thesis supervisor: Wojciech Matusik
Thesis committee: Daniela Rus, Armando Solar-Lezama, Justin Solomon
Abstract:
Designing robots with extreme performance in a given task has long been an exciting research problem attracting attention from robotics, graphics, and artificial intelligence researchers. As a robot is a combination of its hardware and software, an optimal robot requires both an excellent implementation of its body (e.g., morphological, topological, and geometrical designs) and an outstanding design of its brain (e.g., perception, planning, and control algorithms). While we have seen promising breakthroughs for automating a robot's software design with the surge of deep learning in the past decade, exploration of optimal hardware design is much less automated and is still mainly driven by human experts, a process that is both labor-intensive and error-prone.
This thesis argues that it is time to rethink robot design as a holistic process where a robot's body and brain should be co-optimized jointly and automatically. In this work, we present a computational robot design pipeline with differentiable simulation as a key player. We first demonstrate the concept of computational robot design on a real-world copter whose geometry and controller are co-optimized with a differentiable simulator, resulting in designs that outperform human experts by a substantial margin. Next, we push the boundary of differentiable simulation by developing advanced differentiable simulators for intricate, high-dimensional physics. Contrary to traditional belief, we show that deriving gradients can be both science and art. Finally, we discuss new opportunities unlocked by our advanced differentiable simulator and demonstrate them in an example of modeling and controlling a real-world soft underwater robot. We conclude this thesis by asking open questions in differentiable simulation and envisioning a fully automated computational design pipeline for real-world robots in the future.
Thesis committee: Daniela Rus, Armando Solar-Lezama, Justin Solomon
Abstract:
Designing robots with extreme performance in a given task has long been an exciting research problem attracting attention from robotics, graphics, and artificial intelligence researchers. As a robot is a combination of its hardware and software, an optimal robot requires both an excellent implementation of its body (e.g., morphological, topological, and geometrical designs) and an outstanding design of its brain (e.g., perception, planning, and control algorithms). While we have seen promising breakthroughs for automating a robot's software design with the surge of deep learning in the past decade, exploration of optimal hardware design is much less automated and is still mainly driven by human experts, a process that is both labor-intensive and error-prone.
This thesis argues that it is time to rethink robot design as a holistic process where a robot's body and brain should be co-optimized jointly and automatically. In this work, we present a computational robot design pipeline with differentiable simulation as a key player. We first demonstrate the concept of computational robot design on a real-world copter whose geometry and controller are co-optimized with a differentiable simulator, resulting in designs that outperform human experts by a substantial margin. Next, we push the boundary of differentiable simulation by developing advanced differentiable simulators for intricate, high-dimensional physics. Contrary to traditional belief, we show that deriving gradients can be both science and art. Finally, we discuss new opportunities unlocked by our advanced differentiable simulator and demonstrate them in an example of modeling and controlling a real-world soft underwater robot. We conclude this thesis by asking open questions in differentiable simulation and envisioning a fully automated computational design pipeline for real-world robots in the future.