Toyota has announced the T-HR3, a brand-new, third-generation humanoid robot. It’s 1.5-meter tall, weighs 75 kilograms, and has 32 degrees of torque-controlled freedom plus a pair of 10 fingered hands. At first glance, it appears to be very capable, with excellent balance and coordination, and Toyota has decided to approach autonomy by keeping a human in the loop inside of a sophisticated, immersive “Master Maneuvering System.”
As with most flagship robotics projects from large Japanese companies, Toyota has done a very good job of not telling anyone about it until they’re good and ready, meaning that all we have to go on at the moment is a press release and some basic specs and videos. We’ve got those to share, along with some thoughts on what this robot is all about, below.
From the press release:
Toyota Motor Corporation (Toyota) today revealed T-HR3, the company's third generation humanoid robot. Toyota's latest robotics platform, designed and developed by Toyota's Partner Robot Division, will explore new technologies for safely managing physical interactions between robots and their surroundings, as well as a new remote maneuvering system that mirrors user movements to the robot.
T-HR3 reflects Toyota's broad-based exploration of how advanced technologies can help to meet people's unique mobility needs. T-HR3 represents an evolution from previous generation instrument-playing humanoid robots, which were created to test the precise positioning of joints and pre-programmed movements…
We’re trying to figure out exactly what the “third generation” refers to. There were a couple different versions of Toyota’s musical humanoid partner robots from the 2000s; the first one of these could play the trumpet, and was introduced in 2003:
The second could play the violin, and had more muscular legs and significantly better hair:
There’s also this robot, though, which appears to have a completely different set of legs (and a massive battery pack) to enable much more dynamic movements:
All of these humanoids are on the order of a decade old, or older. More recently, Toyota has been focusing on the HSR (Human Support Robot) as a robotic hardware platform, which (as a more conventional mobile manipulator design) is easier to use and far more practical for both research and basic household tasks. This make us wonder why (presumably a year or two ago) Toyota decided to invest in humanoid robots again.
Toyota’s press release addresses this with a very general statement about how the T-HR3 is designed to be “a platform with capabilities that can safely assist humans in a variety of settings, such as the home, medical facilities, construction sites, disaster-stricken areas and even outer space.” This particular version doesn’t seem like the sort of beefy platform that could handle an industrial environment, much less one that’s post-disaster, in contrast to the humanoid robot that Honda has been working on. Rather, T-HR3 looks decidedly more domestic, which is in keeping with Toyota’s overall mission to “support doctors, caregivers and patients, the elderly, and people with disabilities.”
To make sure its robots operate safely around people, Toyota also put a lot of effort in developing a new torque sensing and actuation system. The new Torque Servo Modules, which the company developed in collaboration with Tamagawa Seiki and Nidec Copal Electronics, are used in both the T-HR3 and the Master Maneuvering System. The modules measure forces on the robot’s joints and convey that information to the human operator using force-feeedback. They allow the T-HR3 to control contact forces safely and accurately, and also help the robot to maintain its balance even if it collides with objects in the environment.
But still, why a humanoid robot instead of a mobile manipulator? It’s certainly true that in theory, a humanoid robot is the ideal design to maximize capability in any environment designed for humans. Tasks like going up and down stairs (or going over door thresholds), carrying objects with two hands, and providing direct assistance to humans are all facilitated by a bipedal design with two arms. In practice, however, a robot like T-HR3 is vastly more complicated to control than something like HSR, especially in the context of long-term, reliable, independent autonomy. This, presumably, is where Toyota’s Master Maneuvering System comes in:
T-HR3 is controlled from a Master Maneuvering System that allows the entire body of the robot to be operated instinctively with wearable controls that map hand, arm and foot movements to the robot, and a head-mounted display that allows the user to see from the robot's perspective. The system's master arms give the operator full range of motion of the robot's corresponding joints and the master foot allows the operator to walk in place in the chair to move the robot forward or laterally. The Self-interference Prevention Technology embedded in T-HR3 operates automatically to ensure the robot and user do not disrupt each other's movements.
The Master Maneuvering System is a clever (if complicated and expensive) way of sidestepping the autonomy problem. In the short term, at least, it’s a way of giving T-HR3 a massive amount of pseudo-autonomy by offloading all of the sensing, processing, motion planning, and manipulation tasks onto a human who only has to be mildly trained.
Long term, a system like this probably isn’t a sustainable way to manage a fleet of humanoid robots. Our guess is that Toyota is hoping to use the MMS as a tool to help enable reliable learning by demonstration, where the robot gains experience as it observes humans doing a variety of tasks through it, and eventually is able to understand how to do those tasks itself, in an approach somewhat similar to what Pieter Abbeel is doing at Embodied Intelligence.
The other big advantage of the MMS is that it provides a backup of sorts for autonomous robots that are operating in semi-structured or unstructured environments. If they get stuck, the robots could call for help, and a human could remotely hop in and teleoperate a solution. Just a few humans in Master Maneuvering Systems would be able to provide this service for a much larger fleet of robots that only run into trouble occasionally, which makes the overall system much faster to deploy since it becomes unnecessary to try to reach 100 percent autonomy: Something like 95 percent, plus occasional remote human intervention, is close enough.
It’s tempting, at the moment, to compare T-HR3 to Boston Dynamics’ robots and not be even a little bit impressed. But keep in mind that BD has focused a lot on dynamic mobility platforms, and that’s not what Toyota is trying to do: Rather, they’re more about in-home robots that can be helpful with a focus on precision interaction, and that includes just enough mobility to get around because there’s no reason to include any more. From what we’ve seen, Boston Dynamics does a little bit of manipulation, but not all that much, and their autonomy doesn’t go much beyond tracking people and avoiding obstacles, and they only seem to do that when absolutely necessary. This is in no way a criticism—we’re simply pointing out that companies that make “humanoid robots” take lots of different approaches and focus on different things, and what Toyota is working on is just as useful and relevant and exciting as what we’re seeing from BD.
So far, we don’t have a lot of detail on what, specifically, Toyota is trying to do with T-HR3. We're hoping to see how well the MMS works in semi-structured environments to enable the robot to do task like clearing a table or folding laundry, and if we're lucky, TRI will get their hands on one or two to mess around with and start publishing research papers that we can get a peek at.
[ Toyota ]