Landing with Time-to-Contact and Ventral Optic Flow Estimates

Dario Izzo, Guido de CroonJournal of Guidance Control and Dynamics 2012 · DOI

TL;DR

This paper proposes a novel, vision-only spacecraft landing strategy inspired by insect flight. By combining ventral optic flow (how fast the ground appears to move) with time-to-contact (estimated time until impact), the system can achieve safe landings with reduced sensor and mass requirements, potentially enabling more efficient backup systems or smaller spacecraft.

What Problem Does This Paper Address?

Traditional spacecraft landing systems often rely on a suite of sensors, including altimeters and radar, to accurately determine the lander's state (position, velocity, etc.) for a safe touchdown. While effective, these systems can be heavy and complex, making them less suitable for backup emergency systems or for very small spacecraft where mass savings are critical. The paper explores how to achieve safe landings using only visual information (ventral optic flow) and proprioceptive sensors (like gyroscopes), mimicking the capabilities of flying insects. A key challenge with using only ventral optic flow is that it doesn't uniquely define the vertical descent profile, potentially leading to inefficient fuel usage or undefined landing conditions. Furthermore, in a purely vertical descent, ventral optic flow becomes zero, providing no useful information.

Methods

The paper introduces a landing strategy that combines two visual cues:

  • Ventral Optic Flow ($\omega$): This measures the translational velocity divided by the height. It's a common cue used by insects for controlling speed and altitude.
  • Time-to-Contact (TTC): This is estimated as the height divided by the vertical velocity. The paper proposes estimating TTC using feature scale expansion in the visual field, which is less dependent on the focus of expansion compared to pure optic flow methods.

The authors investigate two primary descent strategies based on these cues:

  1. Constantly Decreasing Time-to-Contact (cdTTC) with Constant Optic Flow (cOF): This strategy aims to maintain a constant ventral optic flow while ensuring the time-to-contact decreases linearly. This is achieved by controlling the thrust based on these visual estimates.
  2. Exponentially Decreasing Time-to-Contact (edTTC) with Constant Optic Flow (cOF): This strategy aims for an exponentially decreasing time-to-contact, theoretically leading to an asymptotic vertical motion. This approach is explored for potentially better mass optimality.

The paper uses a simplified dynamical model of a planetary landing and simulates these strategies using optimal control problems, comparing their mass expenditure and descent times against unconstrained and single-constraint scenarios.

Datasets

Not discussed in the paper. The authors use a simulated environment based on a simplified dynamical model of a planetary landing.

Key Insights and Ideas

  • Vision-Only Landing is Feasible: The research demonstrates that a safe and mass-efficient landing can be achieved using only visual cues (ventral optic flow and time-to-contact) and proprioceptive sensors, eliminating the need for heavier exteroceptive sensors like radar or lidar.
  • Time-to-Contact is Crucial for Vertical Control: Ventral optic flow alone is insufficient for controlling the vertical dynamics of a landing. Incorporating time-to-contact, particularly a decreasing one, provides the necessary information to manage descent speed and ensure a soft landing.
  • Exponential TTC Offers Mass Savings: While a linearly decreasing time-to-contact strategy is explored, an exponentially decreasing time-to-contact strategy shows a significant improvement in mass optimality, reducing propellant usage compared to the linear approach.
  • Inspiration from Nature Works: The strategy is directly inspired by how flying insects like bees land, highlighting the effectiveness of bio-inspired approaches in robotics and aerospace.

Key Results

  • Mass Penalty Analysis: The paper quantifies the mass penalty associated with different landing strategies. A cdTTC-cOF descent in an Apollo-like scenario resulted in approximately 33% more propellant usage compared to an unconstrained descent.
  • Improved Mass Optimality with edTTC: The edTTC-cOF strategy significantly reduced this mass penalty. In the same scenario, the edTTC-COF descent used approximately 18% more propellant than the unconstrained descent, a notable improvement over the cdTTC-COF approach.
  • Feasibility Demonstrated: Simulations show that control laws based on optic flow and time-to-contact estimates can effectively regulate the descent and achieve trajectories very close to optimal solutions, even when thrust saturation occurs.

Limitations and Future Work

  • Simplified Dynamics: The study uses a simplified dynamical model. Real-world landings involve more complex dynamics, including atmospheric effects and non-linearities.
  • Perfect Sensing Assumption: The simulations often assume perfect knowledge and estimation of optic flow and time-to-contact. In reality, these estimates are subject to noise and errors, which would need to be addressed by robust filtering and estimation techniques.
  • Limited Control Law Analysis: While control laws are proposed to demonstrate feasibility, an in-depth analysis of their asymptotic stability is outside the scope of this article.
  • Future Work: The authors suggest that these results are a promising avenue for developing fully automated landing procedures based on visual cues. Further research could focus on real-world implementation, robust estimation algorithms, and more complex environmental conditions.

Summary updated

Abstract

Many recent studies on autonomous spacecraft landing use computer vision methods to improve the accuracy of the state estimates used for landing. Typically, these studies integrate the vision module with other exteroceptive sensors such as laser or radar altimeters. This is a sensible approach for the main landing system of a large spacecraft. However, for a backup emergency system or for much smaller spacecrafts, a solution entirely based on vision and proprioceptive sensors (e.g. gyros) could lead to significant mass savings. Small flying animals are capable of safe and accurate landings while relying only on proprioceptive and visual information. Since this capability holds a promise of landing safely with limited sensors and processing, it has served as inspiration for recent spacecraft landing studies. The focus of these studies has been on the use of ventral optic flow, a measure of the translational velocity divided by the height. Bees are known to use optic flow for controlling their speed and height, also when landing. In particular, when landing, they keep the ventral optic flow constant. Valette et al. study a control law that implements this strategy, simulating landings on the moon. The disadvantages of the sole use of ventral optic flow for landing are two-fold. First, the vertical dynamics of the lander is left free. The ventral flow can have the same constant value for a trajectory in which the lander ascends while accelerating and a trajectory in which the lander descends while decelerating. Thus, one has to directly or indirectly assume some type of descent profile, for example by introducing a pitch law for the spacecraft. Without the use of additional exteroceptive information to compute an optimal pitch profile, this leads to a considerable expense of propellant and to undefined final low-gate conditions. Second, in the case of a straight vertical landing the ventral flow is zero. In such a case, e.g. in the terminal phase of an asteroid landing scenario, the ventral flow does not provide any information on how to land the spacecraft. Advanced Concepts Team, European Space Agency, [email protected] Exteroceptive sensors observe entities external to the spacecraft, while proprioceptive sensors measure quantities within the spacecraft “body”.
BibTeX
@article{izzo2012landing,
  author = {Dario Izzo and Guido de Croon},
  title = {{Landing with Time-to-Contact and Ventral Optic Flow Estimates}},
  journal = {Journal of Guidance Control and Dynamics},
  year = {2012},
  doi = {10.2514/1.56598},
}
Figure