Trajectory Optimization and Mission Planning for Quadrupedal Robots
Lead Research Organisation:
University of Oxford
Department Name: Engineering Science
Abstract
Context and potential Impact
The research focuses only on four-legged robots (quadrupeds) as these mobility platforms offer the highly appealing capability of traversing challenging terrain and environments designed by and for humans (stairs and steps). Our group, the Dynamic Robotic Systems group (DRS), has expertise in deploying the ANYmal for mapping, navigation or inspection missions in challenging industrial environments.
Following the COVID-19 pandemic I adapted my research project to account for the uncertainty in attending lab and office space. As a result I focused on a project that I could work on mostly from home with a high chance of transferring to the robot without further need of development. This research consists of using factor graphs for motion planning. This framework is typically used in state estimation and thus the appeal lies in finding not only a potentially novel approach with low computation times for motion planning, but also in solving two crucial problems using the same tools. State estimation looks back in time while planning looks into the future. Solving the latter means that one could then move onto using the same tool for two problems and thus increasing the efficiency of these algorithms.
Next I will return to the initial project in which I want to investigate how in global mission planning frameworks we can incorporate a better knowledge of the environment to find better mission plans. Currently the only information processed in the existing pipeline within DRS only accounts for geometric information such as how narrow a passage or flat a terrain is. This is not sufficient for real world deployments as the ground type differs (grass, pebble stones, concrete, ...) as well as the environment is made of static objects such as walls and dynamic objects (doors can be closed or open, a box can be blocking the path on one day but not on another). This information needs to be accounted for to increase the chances of deploying robots such as ANYmal into the real world.
Aims and Objectives
The research project aims at improving the autonomous manipulation capabilities using the robotic platform ANYmal:
1. Development of a trajectory optimization pipeline using factor graphs. This objective is a result of replanning for the lockdown following the COVID-19 pandemic and the following lockdown of the lab and office space.
2. Research and development into
a. (Re-)Planning global missions using the existing solutions (Gaitmesh combined with Recast)
b. Integration of semantic information (door, moveable objects) and other physical information (surface friction, ground type) into the planning algorithm
c. Parameter optimization of the framework
Novelty of the research methodology
The novelty of the above mentioned research lies in
1. New approach of trajectory optimization for legged robots using factor graphs, showed and evaluated on the quadruped ANYmal.
2. New mission capabilities for the robot to plan, and execute loco-manipulation tasks in the context of global missions.
EPSRC's strategy alignment and research area
This project falls within the EPSRC Engineering research area.
Collaborations
The Dynamic Robot Systems group (DRS) is involved in the UKRI/EPSRC ORCA and RAIN Robotics Hubs, and the H2020 European Projects 'THING' and 'Memmo'. This project will benefit from access to this extensive network of UK and European researchers, while collaborations in this context are likely to organically grow.
The research focuses only on four-legged robots (quadrupeds) as these mobility platforms offer the highly appealing capability of traversing challenging terrain and environments designed by and for humans (stairs and steps). Our group, the Dynamic Robotic Systems group (DRS), has expertise in deploying the ANYmal for mapping, navigation or inspection missions in challenging industrial environments.
Following the COVID-19 pandemic I adapted my research project to account for the uncertainty in attending lab and office space. As a result I focused on a project that I could work on mostly from home with a high chance of transferring to the robot without further need of development. This research consists of using factor graphs for motion planning. This framework is typically used in state estimation and thus the appeal lies in finding not only a potentially novel approach with low computation times for motion planning, but also in solving two crucial problems using the same tools. State estimation looks back in time while planning looks into the future. Solving the latter means that one could then move onto using the same tool for two problems and thus increasing the efficiency of these algorithms.
Next I will return to the initial project in which I want to investigate how in global mission planning frameworks we can incorporate a better knowledge of the environment to find better mission plans. Currently the only information processed in the existing pipeline within DRS only accounts for geometric information such as how narrow a passage or flat a terrain is. This is not sufficient for real world deployments as the ground type differs (grass, pebble stones, concrete, ...) as well as the environment is made of static objects such as walls and dynamic objects (doors can be closed or open, a box can be blocking the path on one day but not on another). This information needs to be accounted for to increase the chances of deploying robots such as ANYmal into the real world.
Aims and Objectives
The research project aims at improving the autonomous manipulation capabilities using the robotic platform ANYmal:
1. Development of a trajectory optimization pipeline using factor graphs. This objective is a result of replanning for the lockdown following the COVID-19 pandemic and the following lockdown of the lab and office space.
2. Research and development into
a. (Re-)Planning global missions using the existing solutions (Gaitmesh combined with Recast)
b. Integration of semantic information (door, moveable objects) and other physical information (surface friction, ground type) into the planning algorithm
c. Parameter optimization of the framework
Novelty of the research methodology
The novelty of the above mentioned research lies in
1. New approach of trajectory optimization for legged robots using factor graphs, showed and evaluated on the quadruped ANYmal.
2. New mission capabilities for the robot to plan, and execute loco-manipulation tasks in the context of global missions.
EPSRC's strategy alignment and research area
This project falls within the EPSRC Engineering research area.
Collaborations
The Dynamic Robot Systems group (DRS) is involved in the UKRI/EPSRC ORCA and RAIN Robotics Hubs, and the H2020 European Projects 'THING' and 'Memmo'. This project will benefit from access to this extensive network of UK and European researchers, while collaborations in this context are likely to organically grow.
Organisations
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/R513295/1 | 01/10/2018 | 30/09/2023 | |||
| 2280867 | Studentship | EP/R513295/1 | 01/10/2019 | 15/07/2023 | David Rytz |