Research

Multi-Objective Planning in Robotics

Many problems in robotics seek to simultaneously optimize several competing objectives such as length, risk, and social acceptability of robot paths. A conventional approach is using a single cost function comprised of the weighted sum of the individual objectives. Yet, finding representative weights remains an important challenge. Thus, we study methods for computing a finite set of weight vectors whose optimal solutions approximate the set of all attainable solutions. Moreover, we investigate fundamental limitations of the widely used weighted sum approach and its alternatives.
Please check out our 2022 WAFR paper, 2024 TRO paper as well as our 2024 RA-L paper for more details.

Different trade-offs between trajectory length and discomfort for robot trajectories.

Optimized tours considering average and maximum waiting times.

Optimized Wait-Times in Dynamic Vehicle Routing

Recently, we revisited the Dynamic Vehicle Routing Problem (DVRP) that requires robots to allocate incoming tasks among themselves and find an optimal sequence for each robot.

State-of-the-art approaches only consider average wait times and focus on high-load scenarios where the arrival rate of tasks approaches the limit of what can be handled. We studied the missing policies for moderate-load scenarios, where quality of service can be improved by prioritizing long-waiting tasks. To that end, we introduce a novel cost function that takes the p-norm over accumulated wait times and show its theoretical stability as well as empirical advantages for both mean and 95th percentile wait times.
Please check out our 2023 RA-L paper for more details.

Multi-robot Task Assignment and Pickup & Delivery

Autonomous robots find increasingly widespread deployment in service applications. For instance, in hospitals mobile robot platforms help to reduce the workload for qualified staff. Thus, as part of the EU Horizon 2020 project Harmony we investigate multi-robot task assignment for material transport in hospital environments with requests that appear online and have a fixed time window.

When deployed in dynamic, human-centered environments parts of the environment may temporarily become blocked and which robots may only be able to observe on location. We propose an informed planning approach, which can substantially increase the number of on-time deliveries compared to reactive planning approaches.
Please check out our 2022 CDC paper for more details.

Moreover, we recently studied heterogenous robot fleets. Complex task assignment problems require robots with a wider range of capabilities and skills. Mixed fleets deploying specialized robots allow for highly efficient solutions. However, designing robot fleets poses a challenging combinatorial problem.

Please check out our 2023 MRS paper for more details.

Service robots in hospitals

Reactive Planning

Informed Planning

Different tours for information gathering, guided by user priorities over areas of interest.

Learning Objectives for Team Environmental Monitoring

A common planning problem with application in robotics is the well-known team orienteering problem where a fleet of agents collects rewards by visiting locations. Usually, the rewards are assumed to be known to the agents. Yet, in applications such as environmental monitoring or scene reconstruction, rewards are often subjective and specifying them is challenging.

We propose a framework to learn the unknown preferences of the user by presenting alternative solutions to them, and the user provides a ranking on the proposed alternative solutions. This learning based approach allows for quickly adapt complex objectives for environmental monitoring missions to the requirements of end-users.

For more details see our paper in RA-L 2021 .

Improving User Specifications for Robot Behaviour through Active Preference Learning

An important challenge in human robot interaction (HRI) is enabling non-expert users to specify complex tasks for autonomous robots. We study a framework where users specify constraints on allowable robot movements on a graphical interface, yielding a robot task specification. These constraints include traffic rules for autonomous mobile robots such as areas of avoidance, speed limit zones and directional roads. However, users may not be able to accurately assess the impact of such constraints on the performance of a robot. We show users the robot's behaviour on an interface together with an alternative solution where some constraints might be violated. Users then choose between these alternatives, allowing for the robot to learn about the importance of the constraints.

The framework was tested in a user study with a material transport task in an industrial facility. We showed that nearly all users accept alternative solutions and thus obtain a revised specification through the learning process.

For more details see our IJRR paper Improving User Specifications for Robot Behavior through Active Preference Learning: Framework and Evaluation. [video]

Specifications for robot traffic and different routes, depending on how strict the specification needs to be followed.

Computing user-optimal Control Sets

This research investigates the design of a motion planner that can model user preferences. In detail, we generate a sparse control set for a lattice planner which closely follow the preferences of a user.

Given a set of demonstrated trajectories from a single user, we estimate user preferences based on a weighted sum of trajectory features. Simultaneously, we optimize a set of connections in the lattice of given size for the user cost function. The restricted number of connections allows the construction of a control set with small branching factor, ensuring strong performance during subsequent motion planning. Further, every trajectory in the control set reflects the learned user preference while the sub-optimality due to the size restriction is minimized.

We show that this problem is optimally solved by applying a separation principle: First, we find the best estimate of the user cost function given the data, then an optimal control set is computed given that estimate.

For more details see our 2020 WAFR paper on Learning Control Sets for Lattice Planners from User Preferences.

Page updated

Google Sites

Report abuse