Dunster Physical AI

Picking, packing and placing

One hand for an eggshell, a glass of water or an aluminum block: soft or firm on demand, with vision-based force sensing.

Setting

Shared: people within reach, no contact as part of the task

Objects

fragile items, bottles and glass, mixed objects, food items

Environments

lab

Rate

Lab: more than 30 object types; line rate not yet measured

Inspection in hazardous areas

The routine inspection duties of an electrician's day, under supervision: thermal scans of panels and busbars, meter and gauge readings, visual checks of terminations and enclosures, the walk-through before and after energization.

Setting

Separated: guarding keeps people out

Objects

metal parts

Environments

data center, substation, chemical plant

Presentations

in fixture

Rate

Daily rounds; 2027 to 2028

Switching and manipulation in hazardous areas

Racking breakers in and out, opening and closing panels, operating disconnects, verifying de-energization, applying and removing lockout-tagout, driven by the site's own electrician from a safe room.

Setting

Separated: guarding keeps people out

Objects

metal parts

Environments

data center, substation, factory floor

Presentations

in fixture

Rate

Per switching order; first pilot targeted for 2027

Switching and manipulation in hazardous areas

Turning valves, taking samples and handling batteries or chemicals in rooms people should not stay in, from a safe room; autonomy for the repeated steps follows from 2028.

Setting

Separated: guarding keeps people out

Objects

hazardous chemicals, metal parts

Environments

chemical plant, data center

Presentations

in fixture

Rate

Per task; plant owners from 2028

Pharmacy fulfilment (pick, count, pack, label)

Receive and scan stock in, pick the right drug and strength, count and pack, label, present for the pharmacist's final check; designed to take a photo, a video clip and an AI check at every step.

Setting

Shared: people within reach, no contact as part of the task

Objects

pharmaceuticals, bottles and glass, rigid boxes, bags and sacks

Environments

pharmacy, retail backroom, hospital

Presentations

racked shelves, tote

Rate

Design target: a back room around the clock; the first pilot measures picks per hour, accuracy, uptime and steps saved per shift

Compliant self-adaptive variable-stiffness grippers, the hands on both Dunster robots. One gripper goes from soft to rigid, a 70 times stiffness range (0.1 to 7.2 N/mm) with vision-based force sensing, so the same hand holds an eggshell, a glass of water or an aluminum block: more than 30 objects grasped in the published tests. A matching arm link switches between eight stiffness modes in 0.2 seconds for safe contact with people. Published in Robotica (2023) and the Journal of Mechanisms and Robotics (2024).

Where it stands

Lab prototypes; second-generation gripper tested on fragile objects.

Who it is for

The hands on Dunster's robots, not offered on their own; of interest to end users with fragile, mixed or deformable objects.

Robot class

Shared: people within reach, no contact as part of the task

Safety status

Not yet assessed by the Dunster Physical AI Benchmarking Engine

A remote-operated humanoid for high-voltage rooms, chemical plants, data center electrical rooms and confined spaces. The site's own trained electrician or plant operator is to drive it from a safe room with a headset and hand controllers, outside the arc-flash boundary and away from the chemicals. The robot's job is the touching: racking breakers in and out, opening panels, operating disconnects, verifying de-energization, applying and removing lockout-tagout, turning valves, taking samples, handling batteries or chemicals. Every action is to be photographed, recorded and checked by AI, the same verification designed for the pharmacy robot, to give the site a lockout-tagout trail and the insurer evidence. Teleoperation first (2027); supervised inspection rounds such as thermal scans and meter readings next (2027 to 2028); autonomy for repeated steps, with one operator supervising several robots, from 2028. People stay out of the hazard zone at every stage. Same platform as the pharmacy robot; console included. Prices are draft, October 2026.

Where it stands

Gripper prototypes in lab tests; torso in build; first paid pilot targeted for 2027 in a data center electrical room with an electrical contractor.

Who it is for

Electrical contractors on data center builds first; chemical plant owners next; utilities and grid owners from 2029.

Robot class

Separated: guarding keeps people out

Autonomy

A0 Manual: a person controls every motion

Safety status

Not yet assessed by the Dunster Physical AI Benchmarking Engine

A human-sized torso on wheels with two arms and grippers that change stiffness on demand, so one hand holds a glass vial, a carton or a soft bag. Its job is the walking, reaching and lifting in the back room: scan stock in, pick the right drug and strength, count and pack, label, then present the order with its record for the pharmacist's final check. The pharmacist keeps the final check; the robot prepares and documents. Each step is designed to get a photo, a video clip and an AI check, for robot and human work alike, so each prescription has one audit trail; the camera checks and overnight running are design targets for the torso now in build. Design goal: zero wrong medicines, zero wrong doses. It is designed to work at existing shelves with no remodel and to cover nights and weekends, and the same robot could later staff a central hub that fills for many stores. Prices are draft, October 2026.

Where it stands

Gripper prototypes work in lab tests; torso in build at the Harvard Innovation Labs; first paid pharmacy pilot targeted for late 2027.

Who it is for

Independent and hospital pharmacies first; chains' central-fill hubs later.

Robot class

Shared: people within reach, no contact as part of the task

Autonomy

A2 Attended: runs tasks with an operator in the room

Safety status

Not yet assessed by the Dunster Physical AI Benchmarking Engine

A free public rating at dunster.ai/safety for any physical AI system: robot hardware, a physical AI operating system, or software that runs robots. Describe the robot and where it works, answer 36 questions in 8 parts, each citing a clause of the Dunster Physical AI Safety Principles, and get a score out of 1,000, a risk tier from A (Preferred) to D (Refer) and the gaps to fix, part by part. On 29 of the 36 questions a safeguard verified by test or outside review counts for more than one that is only documented. How the questions are weighted, where each tier begins and how a rating feeds into insurance stay inside Dunster. Once the engine's return channel is live, builders on this marketplace start the rating from their product page and the result lands on their profile by itself; until then they rate at dunster.ai/safety and Dunster attaches the result from the engine's records. Indicative only: not a certification, an insurance quote or an offer of cover.

Where it stands

Safety Rating live and free since October 2026; a pilot of the engine license with insurers and MGAs is targeted for 2027; a live risk feed from robot data is planned for 2028.

Who it is for

Providers of robot hardware, physical AI operating systems and robot software; the operators who deploy them; insurers and MGAs writing robotics and AI liability.