Pepper Robot Review: Specs, Cost, and 2026 Status

What Pepper Actually Is

Pepper is a wheeled humanoid designed for face-to-face interaction rather than heavy work. SoftBank Robotics introduced it in 2014 with Aldebaran Robotics engineering, targeting reception, retail greeting, and research use. It stands about 1.2 m (47 in) tall, weighs roughly 28 kg, and moves on a three-wheel omnidirectional base rather than legs.

The head carries two RGB cameras, a 3D depth sensor, and four directional microphones, and a 10.1-inch touch tablet sits on the chest for menu-driven interactions. The Aldebaran Pepper technical documentation lists 20 degrees of freedom across arms, head, and base, with sonars, lasers, and bumpers around the base for obstacle detection.

Pepper is not a manipulation robot. The hands have five fingers but limited payload — SoftBank rates each hand for grasping objects up to a few hundred grams, which is enough to hold a brochure, not to carry a tray. Buyers evaluating Pepper as a service robot should treat it as an interaction endpoint, not a mobile assistant that fetches things.

NAOqi, Choregraphe, and the Developer Story

Pepper runs NAOqi OS, the same middleware used by the NAO humanoid. Behaviors are authored in Choregraphe, a drag-and-drop visual editor that also exposes Python and C++ boxes for custom logic. This lowers the bar for museum, retail, or classroom deployments where non-programmers configure dialogue trees and animations.

The trade-off is age. NAOqi 2.9 is the current line for Pepper, and the SDK ecosystem is smaller than modern robotics frameworks like ROS 2. Teams doing research on SLAM, manipulation, or modern speech pipelines usually bolt an external computer to Pepper rather than staying inside NAOqi.

Speech recognition in stock NAOqi is dated. Most 2024-2026 deployments route microphone audio to a cloud LLM — typically OpenAI or Azure OpenAI — and stream text back to NAOqi for text-to-speech and gesture. RobotLAB and United Robotics Group both market ChatGPT-integrated variants that follow this pattern.

Retail and Enterprise Case Studies

The most cited Pepper pilots are retail greeters and bank branch concierges. Sprint deployed Pepper units at select US stores in 2016 to greet customers and demonstrate accessories, and HSBC introduced Pepper at its Fifth Avenue flagship branch in New York in 2018, later expanding to Seattle and Beverly Hills.

Softbank Robotics America also placed Pepper in Westfield malls, cruise ships, and the Smithsonian; a Wall Street Journal report on a 2018 Margiotta supermarket trial in Scotland is the widely referenced counter-example, where Pepper was pulled after a week because it could not reliably direct shoppers.

The pattern buyers should draw from these pilots: Pepper works as a scripted greeter that draws attention, answers a narrow set of questions, and hands off to a human. It performs poorly as an unsupervised information kiosk in noisy environments with unpredictable queries. That gap is exactly what modern ChatGPT integrations are trying to close, with mixed results depending on ambient noise and question scope.

Production Pause, 2024 Refresh, and Current Status

In 2021 Reuters reported that SoftBank had paused Pepper production and cut headcount at its Paris-based robotics arm, citing weak demand. Sales and support continued for existing customers, but no new units were built.

In 2022 SoftBank sold a controlling stake in its European robotics business to United Robotics Group (URG), which now owns Aldebaran and continues to sell, service, and refurbish Pepper and NAO. In 2024 URG and Aldebaran announced updated software packages and ChatGPT-integrated configurations, but not a new hardware generation.

As of 2026 buyers cannot order a factory-new Pepper from SoftBank. Availability is through refurbished channels and integrators — primarily United Robotics Group in Europe and RobotLAB in North America. Prospective buyers should confirm parts availability (hands, wheels, batteries) and NAOqi support term in writing before signing.

Cost, Lease, and Total Cost of Ownership

Pepper has never had a single public list price outside Japan. The reference points below are the ones vendors and press have named on the record, and buyers should treat any figure as a starting point for a quote rather than a shelf price.

In Japan SoftBank originally sold Pepper for ¥198,000 (about US$1,900 at 2015 exchange rates) plus a mandatory 3-year cloud and insurance contract totaling roughly ¥1.1 million, per SoftBank's 2015 Pepper for Biz announcement. Outside Japan Pepper has always been sold as a business package. RobotLAB markets Pepper bundles that include hardware, software, content, and support and quotes on request rather than posting a fixed price.

Expect the fully loaded cost — refurbished robot, NAOqi licensing, ChatGPT or third-party dialogue software, custom content, delivery, and one year of support — to run in the mid-five figures USD for a single-unit deployment. Multi-unit fleets attract volume pricing but add integration and content-management costs.

Who Pepper Is Right For in 2026

Pepper is a niche, not a default. It makes sense for organizations that want a recognizable humanoid at a fixed indoor station — a museum entrance, a bank lobby, a trade-show booth, a university lab — and that have someone who can maintain scripted content and restart the robot when it faults.

It is the wrong choice for open-ended concierge work in loud environments, for anything requiring manipulation, and for teams that need a modern ROS 2 research platform. Newer service robots and dedicated kiosk hardware often deliver more reliable interaction for less money in those scenarios.

• Confirm refurbished stock, warranty length, and parts availability from URG or RobotLAB in writing.
• Insist on a NAOqi support commitment (version and end-of-support date) in the contract.
• Pilot in your actual ambient noise, not a quiet showroom.
• Budget content updates as an ongoing line item, not a one-time launch cost.
• Name the on-site staff member responsible for daily restart, charging, and content refresh.