New chip could help tiny robots traverse complex environments

2026-06-23 · Source: MIT News - Artificial intelligence · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, medium

Summary

MIT researchers have developed Gleanmer, a new system-on-a-chip capable of generating detailed 3D maps for tiny, low-power autonomous robots and augmented reality headsets. Published on June 23, 2026, this chip consumes only about 6 milliwatts of power, a fraction of existing systems, making it suitable for battery-limited devices like UAVs inspecting industrial HVAC systems. Gleanmer achieves this efficiency by integrating an algorithm that uses compact Gaussian ellipsoids instead of traditional voxels to represent obstacles and free space. This co-design approach includes a one-pass technique for generating Gaussians from depth images and a novel method for fusing overlapping Gaussians directly, significantly reducing memory and power requirements. The chip demonstrated real-time 3D mapping, using only 2.5 percent of the power of the best existing chip and reducing path planning energy by 80 percent.

Key takeaway

For Robotics Engineers designing autonomous systems, Gleanmer's ultra-low-power 3D mapping capability changes how you approach edge device navigation. You can now implement real-time, detailed environmental mapping on battery-limited platforms, like UAVs or AR headsets, without significant power overhead. This enables more complex, collision-free path planning in constrained environments. Consider integrating hardware-algorithm co-design principles to achieve similar energy efficiency in your next-generation compact robots.

Key insights

Hardware-algorithm co-design enables ultra-low-power, real-time 3D mapping for compact autonomous systems.

Principles

• Representing 3D environments with Gaussians is more memory-efficient than voxels.
• Processing data directly on compact representations reduces memory and power.
• Keeping active data in fast on-chip memory boosts energy efficiency.

Method

The Gleanmer system uses a one-pass algorithm to generate Gaussians from depth images, then fuses overlapping Gaussians directly, all within dedicated on-chip memory.

In practice

• Enable UAVs for industrial HVAC inspection.
• Integrate into lightweight AR headsets.
• Efficiently map complex blueprints for AI.

Topics

• 3D Mapping
• Low-Power Robotics
• System-on-Chip
• Gaussian Occupancy Mapping
• Hardware-Algorithm Co-design
• Autonomous Navigation

Best for: Computer Vision Engineer, Research Scientist, AI Hardware Engineer, Robotics Engineer, AI Scientist

Related on AIssential

• CrossMaps: Confidence-Aware Open-Vocabulary Semantic Mapping for Rover Navigation — CrossMaps enables real-time, confidence-aware, open-vocabulary semantic mapping for rover navigation using a dual-memory architecture.
• AMD-Powered 3D Gaussian Splatting for Autonomous Driving Scenes — Street Gaussians on AMD GPUs enables dynamic 3D scene reconstruction and editing for autonomous driving.
• Next AI Leap: 3D Memory — Zip-Map enables linear-time 3D reconstruction by embedding a dynamically trained MLP within a frozen transformer.
• Scientists create robots smaller than a grain of salt that can think — Autonomous microrobots, smaller than a salt grain, now sense, decide, and move using light-powered electric fields.
• Towards Next-Generation SLAM: A Survey on 3DGS-SLAM Focusing on Performance, Robustness, and Future Directions — 3D Gaussian Splatting offers a new paradigm for high-fidelity, efficient, and robust SLAM systems.
• Unitree G1 LiDAR, SLAM, navigation and control. Dev w/ G1 Humanoid P.2 — Integrating LiDAR and SLAM on a humanoid robot requires careful software stack management and addresses core navigation challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT News - Artificial intelligence.