riccardo

# Is Terminator waiting for General AI or simply better batteries?

Updated: Jan 25, 2021

#robotics #autonomous #engineering #mechatronics

In the future, *Artificial **General Intelligence* or other unpredictable events might allow machines to become smarter than humans. However, even without or before similar singularities, robots might soon become more common in our society, hopefully for the best. Improvements of existing A.I. models and hardware are allowing machine learning, visual recognition, and similar processes to empower robotic systems of tangible skills. Despite those technological advancements, in the immediate future, early adopters (human ones) might still find themselves facing basic problems —if and when *General A.I. *will happen machines might be able to solve their problems by themselves by then, again, hopefully for the best. An example I wanted to discuss and quantify here is the electric battery, an element that just a few years ago seemed to be an almost insuperable problem for the success of applications like electric mobility. Can we assume the level of current batteries is such that they would not constitute an impediment to robots in the next few years? Trying to quantify this discussion could result in a good mathematical and technical exercise, regardless of our ability to predict A.I. singularities and similar.

*I am assuming electric robots. I will refrain from discussing here ideas like on-board mini nuclear generators and storing elements like super-capacitors … I want to start with what we have today and what is largely available. I will also neglect always-plugged-in machines. I will consider Lithium-Ion solutions and performances. Currently available robots may present a mixture of electric and pneumatic actuators, I will limit this discussion to all-electric systems since the energy requirements could be comparable while the argument will be simpler. I will offer the possibility to skip calculations and just focus on numbers.*

We can start by quickly referencing the electric mobility industry, a field where the electricity-storing element seemed to be the main factor preventing main players from taking over their markets. Problems were often identified in batteries’ limited storing capacity and excessive cost, even though the limited charging net, customer acceptance, and a cautious approach by the main manufacturers were probably playing important roles too. It seems now we are at a turning point: batteries are reaching energy capacities above 500 Wh/Litre (Li-ion), and the cost per KWh of batteries is decreasing toward the $150 / KWh mark – possibly $100 / KWh soon. Electric mobility appears now as a sustainable and inevitable argument. Can we say the same for robots? Will they just benefit from those technological improvements generated in other fields and be free to focus on completely different issues?

To provide an idea of what we will be thinking about, the reader can look at the following picture of one of the machines from the company Boston Dynamics. __I have to disclose that this post will not imply in any way its numbers and results should be associated with Boston Dynamics and its products; I only know that company because of publicly available information and I referenced them only for the pictures, useful in giving readers an idea of what we may be discussing here.__

We will try below to figure how much power and energy-hungry might robots be and how that would compare with current batteries’ capabilities. The result we will obtain might suggest that depending to some extent on the application, limits of current robots might be represented by components we sometimes neglect in general and high-level discussions.

__Numbers and calculation (after this brief premise I will offer the possibility to skip calculation)__

While we could consider a bunch of specific robotic applications, this post will try to stay general and give *tools* rather than only numbers. We will quantify three main types of tasks a robot would have to *power*. We will then add them together thinking about a possible mix of activities during an average functioning. Finally, we will compare the result with the capacity of current batteries that the robot could adopt. The activities we will quantify are:

Basic activities (repeating ones, e.g. walking)

Major activities (possibly accidental, e.g. load-lifting, jumping, etc.)

Control (running software and related e.g. A.I. computation, data acquisition & processing, movements’ control, cooling system, etc.)

*The reader who wants to jump to conclusions can go directly to the section “Results”*

*1 - Basic activities*

We will calculate an average walking power and walking energy through the simplified robotic scheme below. The picture represents a simplified lateral view of a no-knee-joint robotic leg [90 cm] sustaining and propelling an average 70 kg body (about 700 N) – robots currently marketed can weight around that number. We will have as reference a possible human walking speed of about 5 km/h.

Each step covers 90 cm [0.9 m], corresponding to about 60 degrees [1.05 radiant] of hip rotation. Assuming it would take about 0.6 seconds to cover one step, that would result in walking speed of about 1.5 m/s, or about 5.4 km/h – which is in line with our human speed reference. Assuming that during an average walk the speed in-between steps varies from 4 km/h to 5.4 km/h, we would have an average acceleration during a step-stroke of about 1.4 m/s in 0.6 sec or 2.3 m/s^2. Applying F = m x a = 70 [kg] x 2.3 [m/s]^2 we would need about **160 N of propelling force** – horizontal force in the picture above. The other vertical **700 N** force is just the equilibrium of the vertical forces (about 70 kg * 9.8 m/s^2).

The equilibrium of momentum at the hip – force components perpendicular to the leg times the distance to the hip - would give us the supporting and propelling torque a possible electric motor at the hip would have to generate:

M = { 700 [N] x sin(30 [deg]) + 160 [N] x cos(30 [deg]) } x 0.9 [m] = 441 Nm

Resulting in a power of: P = M x rad/sec = 441 x 1.05 / 0.6 = 771.75 W. Therefore, the energy required to walk 1 hr would be about **0.77 kWh**.

__Important note__: during the 60-degree rotation of the hip, the required torque, therefore energy, diminishes mainly because of the vertical component decreasing its opposing effect perpendicularly to the leg itself. This would be evident during the second half of the stroke. However, considering that we should also increase the required energy roughly by 1.66 because of the entire efficiency of the electric chain of about 60% (batteries charging/discharging, electric motors …), we can hold-on on any adjustments in this walking activity as if the two effects were to offset each other. This approximation might be better excused when we will show the *distance* between the numbers of the results.

*2 - Major activities*

Using a similar calculation of above, and considering a jump of 1 meter performed in 0.5 seconds, the equation of motion will give us the following:

**Vertical distance covered = 0.5 x acceleration x time^2** = 1 [m] = 0.5 x a [m/s]^2 x 0.5 [sec]^2

We would require an acceleration of about 8 m/s^2, that multiplied by the mass 70 kg through F = m x a, it would yield a required force of 560 N. Being this force vertical, we can directly sum it to the initial 700 N still required just to sustain the machine and obtain a required hip torque M = 567 Nm (same equation of above representing the equilibrium of momentum M at the hip). Finally, considering the force for the jump to be applied in the initial 30 degrees (0.5 radiant) of motion in about 0.3 sec, the required power and energy would be about 567 [N] x 0.5 [rad] / 0.3 [sec] = **0.945 KWh**. This would be the energy to keep jumping obstacles of 1 meter in half a second for a straight 1 hour. Dividing this number by a possible 0.6 efficiency of the electric chain (as mentioned above), we would get about **1.57 kWh**.

*3 - Control*

Power consumption of general desktops and laptops could be around 150 W considering just CPU, auxiliary systems, and GPU – a possible reference for the numbers is linked at the end of the post. If we considered the robot covering 40 working hours per week as an average worker, it would imply an energy consumption of 150 Wh x 40 hr = 6 kWh, easily comparable with a medium to high use of a computer that can range around 10 kWh per week – the reader can check on her/his machine in case it is equipped with an energy meter. Even considering an average of 8 kWh per week, that would be about **1.14 kWh per working day**.

Honestly, I think this is too optimistic. Just consider an action-camera where there are heavy image processing activities (image capturing, image stabilization, autofocus, etc.). The batteries of those devices can last barely 1.5 hours under normal and heavy use. Moreover, a robot would have to run many auxiliary systems and many more sensors with the respective data processing streams. Considering those examples, I think the initial energy consumption for control should be penalized by a factor of 7-10, meaning it may be about **1 kWh per hour** rather than per day. Dividing this number by a possible 0.6 efficiency of the electric chain (as above described), we would get about **1.66 kWh.**

__Results__

Summary for those who skipped calculations: main activities, major activities, and control could require respectively **0.77 kWh**, **1.57 kWh,** and **1.66 kWh** to be performed continuously for 1 hr.

Since the last result we obtained [control] is extremely dependent on the software and the related tasks, even if we considered only the first two activities we would already be above **2 kWh** of required energy (0.77 + 1.57 kWh = 2.34 kWh). Adding the third and highly variable component we could easily be close to** 4 kWh **(+1.66 kWh). However, we should consider a possible mix of activities to be performed during each hour; If we considered 70-30-90% we would obtain:

Total energy required = 0.77 [kWh] x 0.7 [hr] + 1.57 [kWh] x 0.3 [hr] + 1.66 [kWh] x 0.9 [hr] = **2.5 kWh**

How does this number compare with the capacity of a possible battery carried and used by the robot? As anticipated above, current Li-ion batteries could have an energy density of about 500 Wh/Liter. We can think that the robot of the picture above could carry about 6L of battery, resulting in about **3.0 kWh** of available energy. So, our robot would have a possible autonomy slightly north of **1 hr**.

Looking through the website of Boston Dynamics I found that the robo-dog they started selling (picture below) could indeed last for about 1 - 1.5 hr, after which the battery would need to be either swapped or recharged. The size of the robot may be different from the one we considered, however, so would be the numbers involved (weight) and the battery’s dimension and capacity; therefore, we can consider the numbers to be comparable.

__Conclusion__

Numbers seem to have some degrees of coherence.

If that was the case, robots might be where examples like electric vehicles where 10 to 15 years ago. In the near future, autonomous or semi-autonomous machines will not be able to just walk around and perform their activities as usually seen in movies, they will have to worry about swapping their power packs or plugging-in for at least 30 minutes almost hourly. Early adopters [human ones] would have to plan accordingly for their use of the machine. It will be interesting to see how [if] science will improve this aspect: better or different batteries, optimal onboard power generation, different concepts decreasing the impact of batteries ... Of course, our decision to consider only largely available batteries made us exclude solutions that could allow longer autonomy right away.

There is also the possibility early robots will be task-specific. Different activities may have extremely different power consumption and the calculation should be addressed accordingly. However, we should not forget that, as the reader may have perceived before, a big chunk of energy is depleted by the system just for *thinking *and controlling. That is also why readers should be careful in comparing these results with performances of devices like drones; those are not necessarily wrong associations, but we need to be careful about the configurations of the machines we reference.

To conclude, this post may have highlighted that a possibly effective way to evaluate or develop new technology might involve working on innovative capabilities while also securing simpler aspects and characteristics. Those features must not be neglected even if we just want to gain an effective understanding of the situation.

*Please do not hesitate in contacting me in case you have questions, doubts, or suggestions about calculations, concepts or even possible future projects*

*riccardo@m-odi.com*

*Boston Dynamics video:*

*https://www.youtube.com/watch?v=uhND7Mvp3f4*

* *

*Some numbers on power consumption related to desktops and laptops*

*https://www.kompulsa.com/much-power-computers-consume/*

* *

*Main photo tag:*

*https://unsplash.com/@ekrull*