Automating the animation of pedestrian motion

To train autonomous driving systems for critical collision scenarios, the DeepSafe project aims to address the “reality gap” in these simulated scenarios. A crucial component of these scenarios is the realistic behavior of pedestrians interacting with their environment. AVSandbox uses states-of-the-art computational biomechanical modelling to generate realistic human motion to meet user-defined pedestrian trajectories. These trajectories are intended to cover more than simply where the pedestrian should walk but to also eventually allow us to describe and include other gestures, such as waving, in these simulations. To make these scenarios as flexible as possible, the motion is automatically generated, based on biomechanical principles, negating the need for motion-capture techniques.

Realistic gait

Human gait speed is not only determined by how fast your limbs are moving, but the range of motion of your joints. If you want to walk faster, it can be more efficient to lengthen your stride than to just speed up your limb movements. By aiming to minimise the energy used whilst walking, we can find the most efficient way to use the limbs to generate the walking speed we require. To maximise the realism of human motion, we use a skeletal model with sophisticated muscle models instead of simpler joint actuators. The muscles are modelled as polylines spanning origin and insertion and include passive and active contractile elements, which are a much closer approximation to reality than joint actuators.

With the use of musculoskeletal models and optimal control techniques, we can impose constraints on high level goals to get realistic movement across a wide range of scenarios. These scenarios include walking, running and traversing of uneven surfaces for healthy people but can also be used to generate more sophisticated motion such as limping or moving with walking aids.  

Video 1: Example gait animation across a variety of walking speeds

Crossing the road

A scenario which is common in many pedestrian cases is crossing a road. This often involves the pedestrian stepping down off a kerb, crossing the road and stepping onto the kerb on the other side. Biomechanically, stepping up and down kerbs is a challenge, involving greater mechanical loads on the knees and an altered gait pattern.

Again, by minimising the energy used, we can find efficient methods of climbing up and down kerbs which result in realistic human movement.

Video 2: Scenario of pedestrian crossing the road


By using computational biomechanical models, and optimisation techniques to solve high-level problems whilst minimising energy usage, it is possible to automate the production of realistic human motion. With these methods AVSandbox has a tool to provide edge cases involving pedestrians to test and train autonomous driving systems in some of the most life-threatening scenarios.

Written by: Erik Meilak, PhDSimulation Engineer

Please get in touch if you have any questions or have got a topic in mind that you would like us to write about. You can submit your questions / topics via: Tech Blog Questions / Topic Suggestion


Got a question? Just fill in this form and send it to us and we'll get back to you shortly.


© Copyright 2010-2024 Claytex Services Ltd All Rights Reserved

Log in with your credentials

Forgot your details?