Robotics
Boltz works on two problems in robotics: designing robotic systems to a cost and performance target, and generating the data robots are trained on.
Inverse design of robotic systems
A robotic system is the robots plus grippers, sensors, safety and layout. These choices affect each other, but they are usually made one at a time. You set targets for cost, speed, footprint and reliability. Boltz returns complete designs that meet them, each checked for reach, collisions and timing.
Machine tending
The target was the fastest design within a fixed budget. Boltz returned a single robot tending four machines. In a second study, a single-robot design beat a far more expensive three-robot design on speed, reach and floor space.
Results from physics-based modeling.
Training data for robots, generated in simulation
Robots learn a new task from demonstrations. Today a person drives the robot through the task again and again. That is slow, it repeats for every new task, and the robot can only become as good as the person. Boltz generates the demonstrations in physics simulation instead. It optimizes the robot's skills for the target class of tasks, under varied operating conditions.
- •Collected by people, one task at a time
- •Limited to what the operator can do
- •Generated across many situations
- •Not limited by the operator's skills
- •Teaches robots the skills to perform their tasks optimally
Contact us
Tell us about the system you need to design or the task a robot needs to learn.