How robots learn to throw objects with precision
In short: Throwing allows robots to reach beyond their physical limits, but it requires mastering complex physics like aerodynamics and friction. This video explores TossingBot, a system that learns to pick and toss arbitrary items into distant targets.
The TossingBot system addresses the difficulty of throwing objects with unknown properties—such as mass distribution, shape, and friction—in unstructured environments. To solve this, the researchers developed an end-to-end formulation that learns to infer grasping and throwing motions directly from visual observations through trial and error.
A key innovation is the use of 'residual physics.' The system combines a physics simulator with deep learning; the simulator predicts base control parameters, while the neural network predicts the 'residuals' or corrections needed to account for real-world dynamics. This synergy allows the robot to achieve over 600 grasps per hour with 85% throwing accuracy, successfully placing objects outside its maximum kinematic reach.
Opens the Intellect+ feed with this card first. Swipe up for the next one.
Sources
- Summary by Intellect · source: This Robot Throws Objects with Amazing Precision
- Also: TossingBot: Learning to Throw Arbitrary Objects with Residual Physics
To cite this card, link to https://intellectplus.co.uk/app/technology-this-robot-throws-objects-with-amazing-precision.
How we source, check and correct cards: Editorial standards.
General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.
Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?
Related
More in Robots · More in Technology · Latest · A–Z · All topics
