The field of edge robotics and embodied intelligence is moving towards more efficient and scalable solutions, with a focus on integrated perception, motion, and communication. Researchers are exploring new approaches to reduce computational complexity, latency, and memory demands, enabling real-time performance on edge platforms. A key direction is the development of models that can form and access memories to stay contextualized in their environment, allowing embodied agents to operate effectively over extended timeframes. Another area of innovation is the design of recurrent sequence models that can integrate visual observations while moving in a scene, enabling spatially aware robots to navigate and situate themselves in previously seen spaces. Notable papers include:

• Learning to Optimize Edge Robotics, which proposes an integrated perception-motion-communication approach that reduces communication overhead and computational complexity.
• Memo, which introduces a transformer-based architecture for training memory-efficient embodied agents with reinforcement learning.
• Kinaema, which presents a recurrent sequence model for memory and pose in motion, enabling robots to navigate and situate themselves in previously seen spaces.