Amazon

The Frontier AI & Robotics team releases a "full-stack" framework designed to bridge the gap between different simulators and real-world deployment.

A New York Times report, based on internal documents, outlines Amazon's strategy to automate three-quarters of its operations, avoiding the need to hire over 600,000 workers. This provides new context for the company's aggressive R&D in humanoid robotics.

Amazon's latest robotics project, ResMimic, uses a two-stage residual learning framework to efficiently teach humanoids complex loco-manipulation skills. By refining a general motion policy with task-specific corrections, the system enables a Unitree G1 to handle heavy and irregular objects with precision.

Amazon's FAR robotics team has unveiled OmniRetarget, a powerful data generation engine that translates human motion into physically realistic trajectories for humanoid robots. The system enables a single human demonstration to be augmented into hours of training data, simplifying reinforcement learning and enabling zero-shot transfer of complex loco-manipulation skills from simulation to a real Unitree G1 robot.

Amazon is reportedly developing AI software for humanoid robots intended for package deliveries, according to The Information. The company is said to be constructing a "humanoid park" in San Francisco for testing, focusing on software while initially using third-party hardware. These reports remain unconfirmed by Reuters and Amazon has not commented.