How Cats Land on Their Feet: Angular Momentum Explained

Biology lens: a reflex that grows up – and its limits

Kittens are not born with a working air-righting reflex. It develops in the first weeks of life and is mature by about 33 days. Cremieux and colleagues compared normal kittens with kittens that had been blind since birth: both developed the reflex at the same pace. The air-righting reflex is therefore mainly controlled by the vestibular system, the balance organ in the inner ear 4.

The tail plays a surprisingly small role. No difference in air-righting has been reported between cats with and without a tail – the turn comes from the spine 1. In geckos it is the other way round: a gecko that falls on its back swings its large tail and turns over in about 106 ms, at the time (2008) the shortest air-righting time reported for animals without wings 1.

The reflex does not make cats safe from falls. Veterinarians call the injuries of cats that fall from buildings high-rise syndrome. Of 132 cats seen over five months, 90 % had injuries to the chest; about 30 % needed no treatment, and 90 % of the treated cats survived 5. A study of 119 cats from Zagreb found an average fall height of four storeys; 96.5 % of the cats survived, but 46.2 % had broken limbs, and falls from the seventh floor or higher caused more severe injuries 6. Landing on its feet spreads the impact over the legs, but it does not cancel it.

Physics lens: angular momentum and a body that changes shape

Angular momentum measures how much a body rotates. For a body turning about a fixed axis, it is the product of its moment of inertia I and its angular velocity ω:

The moment of inertia describes how the mass is distributed around the axis: mass far from the axis counts much more than mass close to it. A cat that stretches its legs has a larger moment of inertia than a cat with its legs tucked in.

Angular momentum changes only if a torque acts on the body. Gravity pulls on every part of the cat, but in free fall it acts as if on the centre of mass and therefore exerts no torque about it. Air resistance is small during the short time the cat needs to turn. So the cat’s angular momentum stays constant – zero, if it is dropped without spin.

How can a body with zero angular momentum turn? For a rigid body it cannot: L = 0 means ω = 0. A cat, however, consists of parts that can rotate against each other. If the front half turns one way, the rear half must turn the other way so that the angular momenta cancel:

The half with the smaller moment of inertia turns more. By tucking in first the front legs and then the hind legs, the cat makes sure that each half turns a lot when it moves forward and only a little when it moves back. After both steps, the whole cat has turned – while L was zero at every moment. Divers and trampolinists use the same effect: they can start twists and somersaults during a jump although their angular momentum does not change 3.

Mathematics lens: how far one cycle turns the cat

Take the two halves of the “tuck and turn” model. A half with tucked legs has the moment of inertia I_t, with stretched legs the larger I_e. The cat twists its halves against each other by an angle φ. With zero total angular momentum, the angle φ is shared in inverse proportion to the moments of inertia:

Step 1 (front legs tucked, twist by +φ): the front half turns by φ · I_e / (I_t + I_e). Step 2 (hind legs tucked, twist back by −φ): the front half turns back by only φ · I_t / (I_t + I_e). What remains after one cycle – for the front and the rear half alike – is

Two results follow directly. If the cat does not change its shape (I_e = I_t), it does not turn at all. And the more the moment of inertia changes, the more each cycle achieves: with I_e = 3 I_t and φ = 180°, one cycle turns the cat by 180° · 2/4 = 90°.

How far does the cat fall meanwhile? In free fall, the height after the time t is

If a turn took 0.3 s – an assumed value for illustration – the cat would drop ½ · 9.81 m/s² · (0.3 s)² ≈ 44 cm before its feet point down. Because h grows with the square of the time, a turn that takes twice as long needs four times the height.

Computer science lens: simulating a turn without spin

Can a body really turn by 180° while its angular momentum is zero at every step? The program below simulates the “tuck and turn” model from the mathematics lens in small time steps. It checks the total angular momentum at every step, counts how many cycles the cat needs to land on its feet and computes how far it falls meanwhile. The moments of inertia and the duration of a twist are assumed values chosen for illustration, not measurements.

import numpy as np
import matplotlib.pyplot as plt

# Two halves of the cat turn about the same axis (the spine), in the "tuck and turn" model.
I_TUCKED = 1.0      # moment of inertia of a half with its legs pulled in (relative units)
I_EXTENDED = 3.0    # the same half with its legs stretched out – assumed 3 times larger
TWIST = 180.0       # how far the front half twists against the rear half in each step (degrees)
STEP_TIME = 0.1     # duration of one twist (s), assumed
G = 9.81            # gravitational acceleration (m/s²)


def twist(i_front, i_rear, d_relative):
    """Twist the halves against each other by d_relative with zero total angular momentum.
    I_f·Δθ_f + I_r·Δθ_r = 0 and Δθ_f − Δθ_r = d_relative give each half's rotation."""
    d_front = d_relative * i_rear / (i_front + i_rear)
    d_rear = -d_relative * i_front / (i_front + i_rear)
    return d_front, d_rear


# One cycle: (1) front legs tucked, twist the front half; (2) rear legs tucked, untwist.
steps = [(I_TUCKED, I_EXTENDED, +TWIST), (I_EXTENDED, I_TUCKED, -TWIST)]
front, rear, time = [0.0], [0.0], [0.0]
largest_momentum = 0.0
for cycle in range(2):
    for i_front, i_rear, d_relative in steps:
        for _ in range(50):                     # each twist in 50 small steps
            d_front, d_rear = twist(i_front, i_rear, d_relative / 50)
            rate_front, rate_rear = d_front / (STEP_TIME / 50), d_rear / (STEP_TIME / 50)
            largest_momentum = max(largest_momentum, abs(i_front * rate_front + i_rear * rate_rear))
            front.append(front[-1] + d_front)
            rear.append(rear[-1] + d_rear)
            time.append(time[-1] + STEP_TIME / 50)

per_cycle = front[100]
print(f"largest total angular momentum during the turn: {largest_momentum:.3f}")
if abs(per_cycle) < 1e-6:
    print("turn per cycle: 0° – without changing shape the cat cannot turn")
else:
    cycles = 180.0 / per_cycle
    fall_time = cycles * 2 * STEP_TIME
    print(f"turn per cycle: {per_cycle:.1f}° (front and rear half alike)")
    print(f"cycles needed to land on its feet (180°): {cycles:.1f}")
    print(f"time needed: {fall_time:.2f} s  →  it falls {0.5 * G * fall_time**2 * 100:.0f} cm meanwhile")

time = np.array(time)
plt.rcParams["font.size"] = 13      # large type: the chart stays readable on a phone
fig, ax = plt.subplots(figsize=(6.4, 5.2), layout="constrained")
for k in range(4):                      # shade the steps in which the front legs are tucked
    if k % 2 == 0:
        ax.axvspan(k * STEP_TIME, (k + 1) * STEP_TIME, color="0.92")
ax.plot(time, front, lw=2.5, label="front half")
ax.plot(time, rear, "--", lw=2.5, label="rear half")
ax.axhline(180, color="gray", lw=1, ls=":")
ax.text(0.005, 186, "feet down (180°)", color="dimgray", fontsize=11)
ax.text(0.205, -52, "grey: front legs\ntucked", color="dimgray", fontsize=11)
ax.set_xlabel("time (s)")
ax.set_ylabel("rotation about the spine (degrees)")
ax.set_title("Changing shape turns the cat – with zero spin")
ax.set_ylim(-60, 230)
ax.legend(loc="lower right")
plt.show()

• Open in Colab
• Download .py
• How to run this code

Output · what the code prints, re-run at every build

largest total angular momentum during the turn: 0.000
turn per cycle: 90.0° (front and rear half alike)
cycles needed to land on its feet (180°): 2.0
time needed: 0.40 s  →  it falls 78 cm meanwhile

What the result teaches:

• Zero angular momentum, 180° turn. The largest total angular momentum during the whole manoeuvre is 0.000 – yet after two cycles both halves have turned by 180°. Each cycle turns the cat by 90.0°, exactly as the formula of the mathematics lens predicts.
• Forward far, back a little. In the chart, each half moves a long way while its own legs are tucked and a short way back while they are stretched. This asymmetry is the whole trick.
• The model is slower than a cat. With the assumed 0.1 s per twist, the model needs 0.40 s and falls about 78 cm. Real cats also bend their body while they twist 1, so they may need less time and height than this model – its numbers are an illustration, not a measurement.

Two programming ideas are worth noticing:

• A conservation law as a test. The program does not just assume that the angular momentum is zero – it computes I_frontω_front + I_rearω_rear at every step and reports the largest value. Checking a physical law inside the simulation catches mistakes early.
• Guarding against division by zero. If the halves never change shape, the turn per cycle is zero and “cycles needed” would be infinite. The program detects this case and says so instead of printing nonsense.

Try it yourself – each change takes one line:

• Set I_EXTENDED = 6.0: a larger change of shape turns the cat by 128.6° per cycle; it needs 1.4 cycles and 0.28 s, and falls only 38 cm.
• Set I_EXTENDED = 1.0: tucking the legs changes nothing – the program reports that without changing shape the cat cannot turn.
• Set TWIST = 90.0: a smaller twist turns the cat by only 45.0° per cycle; it needs 4.0 cycles, 0.80 s and falls 314 cm.

largest total angular momentum during the turn: 0.000
turn per cycle: 128.6° (front and rear half alike)
cycles needed to land on its feet (180°): 1.4
time needed: 0.28 s  →  it falls 38 cm meanwhile

largest total angular momentum during the turn: 0.000
turn per cycle: 0° – without changing shape the cat cannot turn

largest total angular momentum during the turn: 0.000
turn per cycle: 45.0° (front and rear half alike)
cycles needed to land on its feet (180°): 4.0
time needed: 0.80 s  →  it falls 314 cm meanwhile

A real cat has far more joints than two halves. Ara Arabyan and Derliang Tsai built a model of a falling animal with many segments and joints and controlled it in two stages: first turn as fast as possible, then hold the body steady relative to gravity. Their control scheme works out the sequence of movements by itself and includes learning features 7.

Physical AI lens: robots that turn in mid-air

A robot that jumps, falls or floats in space faces the cat’s problem: it has nothing to push against, yet it must land the right way up. Two solutions from biology are being copied.

An active tail. Geckos turn over in the air by swinging their tail; this observation inspired an active tail for a climbing robot 1. Thomas Libby, Robert Full and colleagues showed that leaping lizards swing their tails to redirect angular momentum from the body to the tail. They built a lizard-sized robot with an active tail that used sensory feedback to keep its body level as it drove off a ramp 8.

Legs instead of a tail. Nikita Rudin, Marco Hutter and colleagues trained a neural network with deep reinforcement learning to control the jumping four-legged robot SpaceBok – using only its legs to turn the body in mid-air, as a cat does. The policies were trained in simulation for low-gravity celestial bodies and then transferred to the real robot on a test bed for two-dimensional microgravity experiments, where it jumped and landed again and again 9.

For Physical AI, the falling cat teaches that moving the body can itself be the control. No motor pushes against the ground, no thruster fires – shifting mass at the right moment is enough to change orientation. Learning systems like SpaceBok’s controller discover such movement sequences by trial and error in simulation. A comparison with kittens is only an analogy: their reflex matures within about 33 days 4, but how it develops is a different question from how a robot learns.