Congratulations to Dr Cheston Tan, Senior Principal Scientist, on his involvement in the publication of a review paper in the prestigious journal, Science Robotics.

Dynamics models that predict the effects of physical interactions are essential for planning and control in robotic manipulation. Learning-based dynamics models are created purely from observed interaction data, allowing them to capture complex, hard-to-model factors and uncertainty in predictions, while speeding up simulations that are often too slow for real-time control. Recent successes in this field have shown significant improvements in robot abilities, including long-term manipulation of flexible objects, granular materials, and complex multi-object tasks such as stowing and packing.

The paper “A Review of Learning-Based Dynamics Models for Robotic Manipulation” offers a timely and comprehensive review of current techniques and trade-offs in designing learned dynamics models, highlighting their role in advancing robot capabilities.

Read the full paper here.