Federated Fleets, Tactile Locomotion, and the High-Frequency VLA Stack

High-Frequency VLA Policies for Real-Time Manipulation

This paper pushes VLA models to handle high-frequency manipulation tasks that were previously too fast for vision-language architectures. If you're building dexterous manipulation pipelines, this is basically the missing latency bridge. Worth studying the distillation approach.

Decentralized Memory for Multi-Agent Robot Coordination

Multi-agent coordination with decentralized memory keeps showing up as the practical unlock for warehouse-scale deployment. This one tackles the communication bottleneck head-on. If you're scaling from single-arm to fleet ops, bookmark this.

Tactile-Proprioceptive Fusion for Quadruped Terrain Adaptation

Tactile sensing on quadrupeds for terrain-adaptive locomotion. The interesting bit is how they fuse proprioception with touch to handle deformable surfaces. Legged robots in rough outdoor environments just got a few percentage points more reliable.

Cross-Morphology Locomotion Transfer with Foundation Models

This work tackles how foundation models transfer locomotion policies across morphologies without retraining from scratch. The sim-to-real gap analysis is particularly useful if you're assembling custom robot platforms and need to reuse prior work.

Generalizable Latent World Models for Robot Planning

Latent world models for long-horizon planning keep getting better. This one compresses visual observations into a latent space that actually generalizes across tasks. The practical win is fewer training samples needed per new manipulation domain.

Reachability-Based Safe Control for Collaborative Robots

Constrained safety control for human-robot interaction using reachability analysis. The method is computationally cheap enough to run in the control loop, which matters a lot if you're deploying cobots in shared workspaces. Safety folks should pay attention.

Federated Policy Learning Across Distributed Robot Fleets

Federated learning applied to robot fleets lets you train shared policies without centralizing sensitive data. This is the paper to cite when your customer asks about data sovereignty in multi-site deployments. The communication efficiency gains are real.