Edge Nav, Autonomous Teams, and Data-Driven Safety Guarantees

Research

Robot Team Formation Becomes Autonomous

If your robots can autonomously form teams, they become a system you manage rather than a fleet you micromanage. This is leverage through coordination, not just raw individual capability.

Multimodal Perception Gets a Major Upgrade

Multimodal feature encoders usually struggle with depth or high-res inputs. This improves that fusion, meaning robots can make better multi-sensor decisions.

Tools

Low-Latency Navigation Policies for Edge Deployment

A policy that can run on cheap edge hardware is a policy that can scale. Reducing the compute cost of Navigation factorizes the infrastructure bill for deploying at scale.

Compressing World Models for Real-Time Mapping

Maintaining a world model shouldn't require a cluster of GPUs. This work shows how to compress predictive models enough to run real-time worlds modeling on resource-constrained robots, which is a massive unlock for mobile autonomy.

News

A Real-World Metric for Deformable Object Manipulation

ADBench pulls back the curtain on where Dynamics Generation pipelines actually break down on real-world geometry. Instead of just claiming better generalization, it quantifies it, giving engineering teams a clear metric for improving robustness.

Analysis

Automating Safety Guarantees for Robots

Control Barrier Functions (CBFs) are notoriously brittle; designing them manually limits adaptability. Using data to synthesize CBFs automatically paves the way for safety guarantees that don't break the moment the robot enters a new environment.