- making robots better in the real world
- make robots smarter and think faster
- learning goals for today
- learn some language of robotics (computer architecture)
- understand the important of parallelism
- gain practice in exploring process and opportunities of hardware acceleration
- robotics is a big space
- all robots have sensors, compute, actuators
- robotics pipeline
- planning control algorithm space
- intersection of robotics and adjacent fields
- GPU - computer architecture
- computers aren't really getting faster
- we need to focus on parallelism
- huge boost in speeds by running neural networks
- CPUs and GPUs have fundamentally different strengths and weaknesses
- accelerate rigid body dynamics on a GPU?
- hardware-software co-design
- need to have things work well on CPU and GPU
- rigid body dynamics
- robot physics
- robots are metal
- bunch of different metal joints
- study of forces, F = ma
- algorithmic refactoring is needed to effectively target GPUs
- computations scale better on the GPU leading to as much as 7.6x speedup
- custom hardware can provide more acceleration
- future of robotics - acceleration
- enabling robotics research on high performance parallel architectures
- open-source code, optimized GPU code for the dynamics