Weixuan Zhang

Weixuan Zhang, Lionel Ott, Marco Tognon et al.

The recent development of novel aerial vehicles capable of physically interacting with the environment leads to new applications such as contact-based inspection. These tasks require the robotic system to exchange forces with partially-known environments, which may contain uncertainties including unknown spatially-varying friction properties and discontinuous variations of the surface geometry. Finding a control strategy that is robust against these environmental uncertainties remains an open challenge. This paper presents a learning-based adaptive control strategy for aerial sliding tasks. In particular, the gains of a standard impedance controller are adjusted in real-time by a policy based on the current control signals, proprioceptive measurements, and tactile sensing. This policy is trained in simulation with simplified actuator dynamics in a student-teacher learning setup. The real-world performance of the proposed approach is verified using a tilt-arm omnidirectional flying vehicle. The proposed controller structure combines data-driven and model-based control methods, enabling our approach to successfully transfer directly and without adaptation from simulation to the real platform. Compared to fine-tuned state of the art interaction control methods we achieve reduced tracking error and improved disturbance rejection.

Alexandru Coca, Bo-Hsiang Tseng, Jinghong Chen et al.

Schema-guided dialogue state trackers can generalise to new domains without further training, yet they are sensitive to the writing style of the schemata. Augmenting the training set with human or synthetic schema paraphrases improves the model robustness to these variations but can be either costly or difficult to control. We propose to circumvent these issues by grounding the state tracking model in knowledge-seeking turns collected from the dialogue corpus as well as the schema. Including these turns in prompts during finetuning and inference leads to marked improvements in model robustness, as demonstrated by large average joint goal accuracy and schema sensitivity improvements on SGD and SGD-X.

Weixuan Zhang, Marco Tognon, Lionel Ott et al.

Model-based controllers on real robots require accurate knowledge of the system dynamics to perform optimally. For complex dynamics, first-principles modeling is not sufficiently precise, and data-driven approaches can be leveraged to learn a statistical model from real experiments. However, the efficient and effective data collection for such a data-driven system on real robots is still an open challenge. This paper introduces an optimization problem formulation to find an informative trajectory that allows for efficient data collection and model learning. We present a sampling-based method that computes an approximation of the trajectory that minimizes the prediction uncertainty of the dynamics model. This trajectory is then executed, collecting the data to update the learned model. In experiments we demonstrate the capabilities of our proposed framework when applied to a complex omnidirectional flying vehicle with tiltable rotors. Using our informative trajectories results in models which outperform models obtained from non-informative trajectory by 13.3\% with the same amount of training data. Furthermore, we show that the model learned from informative trajectories generalizes better than the one learned from non-informative trajectories, achieving better tracking performance on different tasks.

Weixuan Zhang, Maximilian Brunner, Lionel Ott et al.

Overactuated omnidirectional flying vehicles are capable of generating force and torque in any direction, which is important for applications such as contact-based industrial inspection. This comes at the price of an increase in model complexity. These vehicles usually have non-negligible, repetitive dynamics that are hard to model, such as the aerodynamic interference between the propellers. This makes it difficult for high-performance trajectory tracking using a model-based controller. This paper presents an approach that combines a data-driven and a first-principle model for the system actuation and uses it to improve the controller. In a first step, the first-principle model errors are learned offline using a Gaussian Process (GP) regressor. At runtime, the first-principle model and the GP regressor are used jointly to obtain control commands. This is formulated as an optimization problem, which avoids ambiguous solutions present in a standard inverse model in overactuated systems, by only using forward models. The approach is validated using a tilt-arm overactuated omnidirectional flying vehicle performing attitude trajectory tracking. The results show that with our proposed method, the attitude trajectory error is reduced by 32% on average as compared to a nominal PID controller.