The evolution of Digit: Agility Robotics’ journey from Cassie to Digit 5

therobotreport.com Sep 17, 2026

The evolution of Digit: Agility Robotics’ journey from Cassie to Digit 5

Curated by Shubham · TechCapsules · Sep 17, 2026

AI-summarised brief · reviewed before publication

Agility Robotics has released Digit 5, its fifth humanoid iteration, after a decade of development that began with the Cassie bipedal robot. The company, now with 400 employees and a factory capable of 10,000 units annually, has evolved from research prototypes to commercial deployments. Digit 5 features human‑like legs, extended battery life, and enhanced strength, designed for logistics and manufacturing environments. Earlier models, such as Digit 3, were sold to Ford and used for package delivery, while Digit 4 focused on warehouse tasks.

💡 Why It Matters

· The shift to human‑like legs and stronger payloads signals a move toward robots that can safely and efficiently integrate into existing industrial workflows, potentially reshaping labor dynamics in logistics and manufacturing.

Capsules Robotics

Read full article on therobotreport.com

Agility’s New Digit 5 Robot Has $300M in Orders — But There’s…

Agility Robotics announced its new Digit 5 humanoid robot, standing 5‑foot‑11 and weighing 284 pounds, designed to work beside human workers in factories and warehouses. The company reports over $300 million in multi‑year orders, largely from a single customer agreement, with first‑access slated for early 2027 and general availability later that year. Digit 5 replaces the previous model’s awkward knees with forward‑facing legs powered by cycloidal actuators, enabling it to squat, lift 50 pounds, and operate for more than 20 hours daily thanks to a rapid 9‑minute charge cycle. Safety features include multimodal sensors, independent controllers, and a self‑protective seated posture when an obstacle appears. Despite the strong order book, Agility remains loss‑making, with 2025 net sales of $1.78 million against a $140.2 million operating loss, and faces growing competition from Boston Dynamics, Tesla, and Figure AI.

💡 The robot’s ability to work unencumbered beside humans could redefine labor dynamics in high‑volume manufacturing, potentially reducing reliance on costly automation barriers. Its rapid charging and safety systems address key operational hurdles that have stalled wider humanoid deployment.

engadget.com · Sep 17

Japanese company develops a humanoid robot ‘ambulance’ for on-the-spot repairs

Japanese firm GMO Internet Group has launched Japan’s first “humanoid ambulance,” a specially equipped maintenance van that roams Tokyo delivering on‑site repairs to its own fleet of service robots. The vehicle carries diagnostic tools, spare parts, engineers and a backup humanoid that can replace a malfunctioning unit while the broken robot is taken to GMO’s Humanoid Lab in Shibuya. Only one ambulance is currently deployed and it serves robots operated by GMO AI & Robotics Trading, including baggage‑handling units at Haneda Airport that run on Unitree hardware. Designer Yasumichi Morita styled the van with purple lightbars and the label “Humanoid Ambulance.” GMO’s CEO says the service prevents downtime after earlier incidents required off‑site repairs for future widespread robot deployment worldwide.

💡 By pre‑positioning repair capacity, GMO turns robot downtime into a manageable service issue, accelerating adoption of autonomous workers in dense urban settings.

therobotreport.com · Sep 16

OLogic to share how robots can learn from human demonstrations at RoboBusiness

OLogic will present at RoboBusiness 2026 on egocentric robot learning, a method where robots acquire skills by observing human demonstrations captured via wearable cameras and teleoperation. The session, led by CEO Ted Larson, will cover the technical foundations, challenges, and future implications of this approach, which has attracted attention from NVIDIA, Google DeepMind, Meta, and emerging robotics startups. By enabling robots to learn directly from human behavior, the technology promises faster, cheaper deployment across manufacturing, logistics, healthcare, and service robotics.

💡 The talk highlights a shift toward human‑centric training that could dramatically lower the barrier to entry for advanced automation, reshaping how industries adopt robotics.

spectrum.ieee.org · Sep 16

Rethinking Robot Safety in the Age of AI

Researchers are exposing critical safety gaps in modern “Physical AI” robots that rely on multimodal sensors and AI-driven decision making. Studies such as BadVLA and GoBA demonstrate backdoor attacks that trigger malicious action trajectories when specific visual cues appear, achieving up to 97 % success without degrading normal performance. System‑level flaws compound the risk: the UniPwn Bluetooth exploit disclosed in September 2025 can hijack quadruped and humanoid robots, decrypt traffic, bypass authentication and inject commands, potentially spreading across fleets. Vulnerabilities in ROS 2 and DDS middleware further allow arbitrary code execution or unauthorized model weight changes, undermining trust in command streams. Simulation tools like NVIDIA Isaac Sim paired with VicOne’s Radeis validator aim to test adversarial inputs before deployment, highlighting a shift toward proactive safety validation in robotics.

💡 Hidden triggers let compromised robots act normally until a specific cue appears, turning everyday objects into covert weapons. Demonstrated exploits show entire robot fleets can be commandeered in minutes, demanding new validation and monitoring standards.

en.sedaily.com · Sep 16

UNIST Develops Ultrafast Spatial AI That Cuts Mapping Errors 45.9%

Researchers at Ulsan National Institute of Science and Technology have created UniSim‑SLAM, an AI‑driven spatial recognition system that builds precise three‑dimensional maps from camera images while tracking moving objects in real time. The method merges rapid two‑frame pose estimation with detailed sub‑map analysis, using Sim(3) factor‑graph optimization to align scale, rotation and position across both data streams. Benchmarked on the 7‑Scenes indoor dataset, UniSim‑SLAM achieved an average path error of 2.0 cm, cutting errors by up to 45.9 % versus the prior state of the art, and processed images in 0.197 seconds—over 17 times faster than VGGT‑SLAM. The algorithm retained accuracy with different position‑tracking models, indicating broad applicability for autonomous vehicles, service robots and AR devices. It promises reliable navigation in dynamic settings.

💡 By slashing drift and latency, UniSim‑SLAM makes high‑speed indoor navigation feasible for robots and AR, where even centimeter‑scale errors can cause collisions or misalignment.

forbes.com · Sep 15

The Denominator Problem: What Warehouse Robots Taught Me About Reliability

Shivansh Inamdar, a robotics engineer at Nimble Robotics, explains how a 95% pick‑accuracy rate in warehouse robots translates into thousands of daily failures when scaled to 50,000 picks. He argues that commercial viability hinges not on eliminating failures but on designing systems that manage them efficiently. Inamdar highlights that each percentage point improvement reduces failures by hundreds, yet even a 97% success rate still yields 1,500 errors. The article underscores the importance of robust error handling in large‑scale autonomous operations.

💡 The piece reveals that reliability at scale demands proactive failure management, a principle applicable beyond robotics to any high‑volume automated system.