Agility Robotics unveils Digit 5 humanoid with AI software stack for autonomous warehouse work

On September 15, 2026, Agility Robotics announced Digit 5 from its Salem, Oregon headquarters. The announcement (press release, product page, and executive commentary from CEO Peggy Johnson and co-founder/Chief Robot Officer Jonathan Hurst) presented:

• A new safety architecture intended to eliminate physical safety barriers: proprietary AI-based human detection across a 360° radius using multiple vision-based sensors; visual and auditory safety cues communicating motion intent; and an independent safety controller that triggers protective responses without waiting on the main compute stack. If a person gets too close, Digit 5 can autonomously avoid, stop, or lower itself into a statically stable seated position and shut down its motors before a person can touch it.
• Substantial hardware upgrades over Digit 4: new human-like (non-bird-leg) leg design with proprietary cycloidal actuators enabling repeated 50 lb (22.7 kg) lifts (+40% payload; covers OSHA single-person lift tasks); 90-minute battery that recharges in 9 minutes (10:1 run-to-charge ratio vs 2:1, enabling >20 productive hours per 24-hour day); swappable end-effectors on ISO-standard mounting flanges; taller reach to 7.2 ft (2.2 m) vs 5.5 ft; 5'11" (1.81 m) tall, 284 lb (129 kg), 30 motors.
• Software/workflow expansion: from Digit 4's tote-handling focus to full-facility workflow coverage — depalletizing inbound goods, machine tending, kitting and sequencing, quality inspection, and palletizing outbound — orchestrated by the four-layer physical-AI stack (hardware foundation → 1 kHz motor-coordination layer → skill-definition layer trained via teleoperation, mocap and physical RL → LLM/vision-language-action cognitive orchestration layer) and coordinated with existing AMRs, conveyors and warehouse systems by the Agility Arc cloud platform (uptime, throughput, MTBI KPIs; WMS/WES/MES integration).
• Commercial claims: more than $300 million in multi-year customer orders for Digit 5 as of May 2026 (subject to contractual milestones), a pipeline of 30+ prospective customers, and first planned international expansion (EU and UK, CE mark expected). Early access H1 2027; general availability end of 2027.
• Safety milestone claim: describing itself as the first humanoid to pass an independent field evaluation on a customer production line against industrial safety standards administered by OSHA (company claim; see Risks).
• NVIDIA partnership: Digit 5 is the first launch partner for NVIDIA's Halos for Robotics platform, pairing Agility systems with NVIDIA IGX Thor compute and Halos Core safety infrastructure, plus the Halos AI Systems Inspection Lab for certification work (partnership pre-announced June 22, 2026; Digit 5 is the first production robot shipping with NVIDIA Halos OS per NVIDIA).

Context: the launch lands as Agility prepares to go public via a June 2026 SPAC merger with Churchill Capital Corp XI ($2.5B pre-money, >$620M expected gross proceeds, ticker AGLT, expected Q4 2026 close), with the order book being effectively a single ~1,000-robot three-year RaaS contract with warrants (per SEC materials) — an important caveat for the "orders" headline.

• Safety paradigm for commercial humanoids: the industry's default — isolated, roped-off or caged robot workcells — is challenged by a production-intent humanoid whose safety case rests on onboard perception + an independent safety controller rather than external fencing. Agility positions this as moving humanoids "from pilots to industrial deployments at scale."
• Capability envelope: payload up 40% (50 lb repeated), reach up 31% (7.2 ft), run-to-charge ratio up from 2:1 to 10:1 (>20 hrs/day), human-like legs replacing backwards knees, swappable end-effectors, and workflow scope expanded from tote-handling to full receiving→shipping workflows.
• Software stack emphasis: the "AI software stack" (physical-AI training, LLM/VLA orchestration, Arc fleet management) is now the headline differentiator — humanoid value is shifting from hardware to embodied-AI software and fleet platforms.
• Go-to-market: first international expansion (EU/UK 2027), RaaS pricing ($8,500/month) alongside outright purchase ($200K), and public-market financing to scale production at RoboFab (10,000 units/year target capacity).

• Before: Digit 4 operates inside physical safety barriers/cages in a human-free zone at customer sites (GXO Flowery Branch, Schaeffler, Amazon, Toyota Motor Manufacturing Canada, Mercado Libre); ~65,000 cumulative hours; focused on tote-based material handling; 35 lb payload, 5.5 ft reach, 2:1 run-to-charge; backwards-knee legs; single-purpose workcells; pilots and early commercial deployments.
• Change (2026-09-15): Digit 5 unveiled — engineered from the ground up for cooperative safety (avoid/stop/sit, independent safety controller, 360° human detection), 50 lb payload, 7.2 ft reach, 10:1 run-to-charge, swappable end-effectors, full-workflow autonomy from depalletizing to palletizing, orchestrated by a four-layer physical-AI stack and the Arc fleet platform, on NVIDIA IGX Thor + Halos.
• After: Humanoids able to share open factory/warehouse floor space with humans without cages; one robot covering end-to-end workflows that previously required specialized automation at each stage; order book of $300M+ (≈1,000 robots RaaS per SEC materials) to fulfill from RoboFab; EU/UK entry planned for 2027; public listing via SPAC (AGLT) expected to fund production scale-up; stands as the de-facto commercial benchmark other humanoid vendors (Figure, Apptronik, Unitree, Tesla Optimus) must match on safety.

Safety architecture (Agility's stated design):

1. Safe human detection — proprietary AI algorithms + multiple vision-based sensor types (CTO Pras Velagapudi: "a few different types of vision-based sensors") continuously monitor 360° around the robot; risk is evaluated dynamically with no predefined zones — the robot slows, then stops, avoiding as approach distance decreases.
2. Safety cues — visual and auditory signals convey motion intent so nearby humans can participate in safe operation.
3. Safe motion control — an independent safety controller (separate from the main AI compute) oversees detection of humans within unsafe distance and instantly triggers protective responses.
4. Seated failsafe — as a person enters immediate proximity, Digit 5 lowers itself to a statically stable seated position on the ground and turns off its motors ("the robot needs to be in a statically stable posture and be able to turn off its motors before a person can touch it" — Hurst). This directly addresses the core humanoid risk: a 284 lb biped falling on a person.

Four-layer physical-AI stack (Hurst, per briefing to Humanoids Daily):

1. Hardware Foundation — custom structure, cycloidal actuators, mechanics for repetitive dynamic loads, 7.2 ft reach.
2. Coordination Layer — low-level motor control across 30 onboard motors at 1 kHz; dynamic bipedal gait stabilization, balance, reflexive reaching/walking primitives.
3. Skill Definition Layer — domain-specific manipulation actions bootstrapped via teleoperation, motion capture, and simulated kinematic animation, then refined through physical reinforcement learning.
4. Cognitive Orchestration Layer — LLMs and vision-language-action reasoning interpret human task descriptions and assemble defined physical skills into multi-step workflows (e.g., depalletize → machine tend → kit → inspect → palletize).

Compute/safety platform: NVIDIA IGX Thor (industrial-grade AI compute, dedicated Functional Safety Island) + Halos Core (safety OS running Linux/QNX) + Halos AI Systems Inspection Lab for certification preparation — Agility is the first launch partner; Digit is the first production robot shipping with NVIDIA Halos OS.

Fleet layer: Agility Arc cloud platform provides fleet KPIs (uptime, throughput, Mean Time Between Incidents), over-the-air management, and integration with WMS/WES/MES; coordinates handoffs with AMRs and conveyors. Data: diagnostics go to Arc encrypted; camera/depth sensor data processes locally.

Battery/economics: 90-minute runtime + 9-minute autonomous charge → 10:1 run-to-charge; autonomy docks itself and recharges between tasks; designed to sustain >20 productive hours/day.

• Safety is the bottleneck being removed. The dominant constraint on humanoid commercial deployment has been the cage: every prior Digit ran inside fenced workcells. Digit 5 is a credible, production-intent attempt to make "cooperatively safe" operation verifiable — the difference between humanoids as point automation (isolated cells) and humanoids as general-purpose floor workers (shared space).
• First-mover commercial evidence: nearly three years of continuous commercial deployment and 65,000+ operational hours give Agility a data moat competitors lack; the $300M+ order book (≈1,000 robots per SEC filings) is the largest disclosed humanoid commitment to date, even with milestone caveats.
• Defines the benchmark for the humanoid race: Figure, Apptronik, Unitree, Tesla and others are racing toward customer shipments; Digit 5 sets the safety/autonomy/economic target they must match — including a concrete safety response (avoid/stop/sit), RaaS economics, and a physical-AI software story.
• Physical AI stack as the new product: the announcement frames software (skill training, LLM orchestration, Arc) as the core of the platform — aligning Agility with the "AI-native humanoid" thesis and the broader week's theme of autonomous agents moving into the physical economy (cf. Salesforce AIforce + Siemens, NVIDIA's Jensen Huang "agents will program themselves" remarks at Dreamforce, S26/S38).
• Capital-markets validation: the SPAC (AGLT) will make Agility the first U.S.-listed pure-play humanoid company — a public-company test of humanoid economics (2025 net sales were only ~$1.8M per SEC filings, i.e., a revenue story yet to be proven at scale).

• Unfenced humanoid deployment in warehouses/factories: robots sharing aisles, docks and workstations with people, without costly perimeter interlocks and plexiglass cells.
• A single general-purpose robot covering an entire facility workflow — depalletizing → machine tending → kitting/sequencing → quality inspection → palletizing — replacing chains of specialized automation.
• Continuous near-24h operation via 9-minute charging (>20 productive hours/day), removing the shift/battery constraint on robot ROI.
• Fleet-scale management through Arc (uptime/throughput/MTBI dashboards, WMS/WES/MES integration), making multi-site fleets operationally plausible.
• International adoption: EU and UK entry (CE mark path) and ANSI/ISO standards contributions position cooperative-safety humanoids for regulatory frameworks rather than bespoke approvals.
• At $8,500/month RaaS, employers can displace roughly a full-time-equivalent of $30.50/hr loaded labor cost (≈$100K/year per robot avoidance per Agility's illustrative math, per IEEE/SEC materials) — making the economic case testable now, not in five years.

1. Who is the anchor customer behind the ~1,000-robot order, and what milestones gate conversion?
2. What exactly did the OSHA-standard field evaluation test, and can it be replicated by independent auditors?
3. Which sensor types and perception latencies power the 360° human detection, and can the seated failsafe fire within safe stopping distances for a 284 lb biped carrying 50 lb?
4. What SIL/PL certification level is Digit 5 targeting, and on what timeline relative to ISO 25785-1 finalization?
5. What is real-world sustained throughput (picks/pallets per hour) vs the 20-hour theoretical utilization?
6. Will Agility publish Arc/digital-twin APIs for third-party skill development, or keep skills closed?
7. How will Agility's public-company financials alter RaaS pricing once deployment costs are visible?
8. What happens when the EU AI Act machinery (CE, AI Act risk classification) applies to humanoids in EU facilities in 2027?
9. Does cooperative safety extend beyond warehousing to the home market (per CEO letter "opening a path into broader markets over time — including the home"), and on what safety evidence?
10. How quickly will rivals (Figure, Apptronik, Unitree G1+, Tesla Optimus) match the avoid/stop/sit safety pattern and compress pricing?

• Q4 2026 (near-term): closure of the Churchill Capital Corp XI SPAC (shareholder vote, S-4 effectiveness, HSR/regulatory clearances, minimum $200M at closing); Agility begins trading as AGLT; RoboFab retooling for Digit 5 production ramps; milestone-gated order conversion begins.
• H1 2027: Digit 5 early-access deployments start; first sustained human-adjacent operating data becomes available; expect first public throughput/MTBI figures.
• End of 2027: GA for manufacturing, warehouse and distribution operators; EU/UK commercial entry with CE marking; ISO 25785-1 work progresses toward a standard, potentially reshaping certification requirements mid-launch.
• 2027+ (predictions): rivals ship comparable safety patterns within 12–18 months; RaaS pricing pressure intensifies as Unitree-class hardware undercuts cost; Agility's innovation engine shifts from hardware to skill breadth and fleet software; a significant fraction of the anchor order converts only if field safety data holds, making H1 2027–2028 field reports the true test of the $300M story.

Digit 5 is the most commercially serious statement yet that humanoids are entering the physical economy as cage-free workers — not as demos. The announcement matters less for the spec sheet (50 lb payload, 9-minute charging, 20-hour days) than for what it signals: safety is now a product-architecture problem the industry can solve, and Agility has converted three years of real warehouse operations into an order book, a standards position, and a public-company vehicle. The healthy skepticism belongs exactly where independent analysis points it: the $300M order book is one contract with warrants and milestones, the OSHA evaluation is a company claim, and "cooperatively safe" will only be proven by years of field data. In a news week dominated by software agents, Digit 5 is the reminder that the most consequential agents of all are the ones with legs — and that boring reliability, not flashy demos, is what gets them hired.

Evidence discipline summary: FACT (announcement date, specs as announced, SPAC terms, SEC order-book framing); COMPANY CLAIM (OSHA field-evaluation milestone, cooperative-safety efficacy, $100K/year savings illustration, end-to-end workflow readiness); INDEPENDENT EVIDENCE (multi-outlet same-day coverage, NVIDIA Halos partnership pre-announcement, SEC filings analysis of order concentration/warrants, analyst caution on unfinished standards, Ars/IEEE technical assessment of the seated-safety design). No rumors used.