It has a humanoid shape but can't stand straight! What's wrong with humanoid robots?

Why can't common humanoid robots stand upright?

Author: Xing Boyang

With the rapid development of science and technology, humanoid robots have become a part of our lives. From home service robots to industrial robots to performance robots in the entertainment field, they are everywhere. However, have you ever thought about this question: Why can't these seemingly highly advanced humanoid robots always stand straight?

First, we need to understand the structure and movement of humanoid robots. Humanoid robots are usually composed of a head, a torso, limbs, and joints, which are moved by motors and reducers. When walking, humanoid robots need to adjust their body posture by rotating their joints to maintain balance. However, this process is not that simple.

So why can't humanoid robots stand straight? There are several reasons for this:

1. Mechanical structure limitations

Mechanical structure is one of the key factors that affect the inability of humanoid robots to stand upright. Humanoid robots have a relatively complex structure and many joints, which makes them more prone to friction and resistance during movement. Especially at the joints, due to the existence of friction, it is difficult for the robot to maintain an upright posture while walking. In addition, the weight distribution of the humanoid robot is also an important factor. If the weight is concentrated on one side, it will cause the robot to tilt while walking.

2. Problems with motion control

From the perspective of robot control, knee bending can ensure that the robot's posture is non-singular, thereby increasing the robot's controllability. Singular postures are situations such as straightening the knee joint and coaxial joints. In these situations, either the accessible space of the joint becomes smaller (the straight knee causes the ankle to move only in an arc with the leg length as the radius, and cannot move to the area within the arc), or a degree of freedom is lost, resulting in a more extreme force situation (for example, after the knee is straightened, the load-bearing performance in the direction perpendicular to the leg is extremely poor, but the load-bearing capacity along the leg direction is doubled). In short, it will add difficult-to-handle tricky situations to the system control. Similar to dividing by O in mathematics, this happens frequently, resulting in the addition of many additional means in numerical calculations to prevent the occurrence of "division by O" during the calculation process.

The advantage of walking with bent knees is that the robot's posture control is more controllable. Because the legs can still be straightened, the legs have enough adjustment space to exert control when the upper body adapts to disturbances caused by terrain undulations. For example, before a swing leg lands, if the actual center of gravity of the robot's upper body is higher than expected due to unexpected shaking, the leg landing step can reduce the knee bending angle and make the legs longer to compensate for the floating center of gravity error.

In layman's terms, if we walk with bent legs and our body sways, we can straighten our legs a little to adjust our posture. If we keep walking with straight legs, we don't have the option of straightening our legs more, and there is less room for adjustment.

Humanoid robots consume a lot of energy when walking. The subtlety of human gait energy saving is that it is an energy-saving movement that is not completely stable. When people walk, they seem to keep using sticks to support their bodies and swing forward, and then keep removing the sticks behind. In the process of walking, most of the weight is introduced to the ground by the support force of the stick along the axis, and the force of the joints is only used to drive the forward swing to maintain forward movement, and to perform short braking when switching gaits to prevent tipping.

The incomplete stability of human gait lies in the fact that walking may indeed lead to falls. Because people walk with straight legs, there must be a moment in the gait when one leg is straightened, and the body is supported by the thigh and calf bones being approximately coaxial to reach the dead point of the connecting rod. At this time, the knee joint of the straightened supporting leg has almost no driving torque. At this time, if the body is shaken due to various reasons (such as sudden change of direction or emergency stop), if the strength of the hip and ankle joints cannot fully provide the anti-tilting torque, the body posture will tend to tilt. The supporting leg does not have much room for movement (the straight knee joint causes the hip joint to only move around the ankle on a cone surface and cannot move along the leg), and then the body posture will gradually become unstable and cause a fall (if other limbs do not take measures)

Landing on the sole of the foot with straight legs is like a stick extending from the body. The force transmitted to the body by straight legs is mainly thrust along the leg axis and torque of the hip and ankle joints. The force transmission direction is limited, so the landing position must be calculated very carefully each time, otherwise the body will lose balance after a single landing. Walking with bent legs can adjust the size and direction of the force transmitted to the body by the landing leg by bending the knee joint.

This is why when people first try to walk on slippery ice, they will subconsciously bend their knees to ensure their body balance, rather than just stretching their legs out and stepping on it, which will most likely cause them to slip or split.

At present, humanoid robots can achieve a relatively large posture adjustment space in terms of control, which has a positive effect on stabilizing the upper body posture and gait. However, non-humanoid gait will make the robot consume a lot of energy when walking, which makes the power estimation in system design too large, resulting in larger drive joints, making the robot bulky.

3. Inadequate sensor performance

In order to achieve precise motion control, humanoid robots need to rely on various sensors to obtain environmental information. Currently, most humanoid robots on the market use sensors such as lidar, cameras, gyroscopes, and accelerometers. However, the performance of these sensors is limited and it is difficult to fully capture changes in the environment. Especially when the light changes greatly or there are debris on the ground, the sensors may misjudge, causing the robot to be unable to accurately judge its position and posture, and thus unable to maintain a stable standing state.

4. Artificial intelligence algorithm optimization

With the development of artificial intelligence technology, more and more researchers have begun to try to apply deep learning, reinforcement learning and other methods to the field of humanoid robots. These algorithms can help robots better understand and adapt to environmental changes, thereby achieving more stable and natural walking.

However, applying these advanced algorithms to practical problems still faces many challenges, such as insufficient data and high algorithm complexity. Therefore, how to combine these algorithms with existing motion control methods to improve the stability and walking performance of humanoid robots remains an urgent problem to be solved.

In short, the current phenomenon of humanoid robots not being able to stand straight is the result of a combination of factors. In the future development process, we need to continuously optimize mechanical structure, motion control algorithm, sensor performance and other aspects, and use artificial intelligence technology to improve the robot's stability and walking performance.