Learning Robot Policies from Sparse Success Signals
This paper introduces STL-Guided Stein Variational Policy Gradient (STL-SVPG) for learning robot policies in tasks with sparse success signals.
IAEI / RESEARCH & IDEAS
Research briefs on embodied intelligence, robotics and learning through interaction, with links to original sources.
FOLLOW THE RESEARCH
Briefs are automatically prepared from public research abstracts. They do not imply independent verification or association endorsement.
This paper introduces STL-Guided Stein Variational Policy Gradient (STL-SVPG) for learning robot policies in tasks with sparse success signals.
This paper introduces RIVET, a framework that uses a shared object-centric representation to generate and integrate executable programs for robotic manipulation tasks.
The paper introduces ReVAMP, a vector-accelerated motion planning method using reparameterization through analytic inverse kinematics, which aims to improve planning speed for robots with complex constraints.
Rolling-WAM addresses latency issues in World Action Models by distributing denoising across planning cycles, enhancing robotic manipulation performance and speed.
This paper introduces RAPID, a system that automatically generates, verifies, and refines robot programs from a single visual demonstration.
The World Action Agent (WAA) uses vision-language models (VLMs) to directly control robots in a visual action workspace, enhancing decision-making and skill acquisition.
RoboTwin-Phys addresses the lack of physical condition diversity in robot manipulation benchmarks by introducing a new benchmark that varies physical parameters like mass, friction, and joint dynamics.