Skip to main content

subTerranean Haptic INvestiGator

Periodic Reporting for period 1 - THING (subTerranean Haptic INvestiGator)

Reporting period: 2018-01-01 to 2019-06-30

The goal of THING is to advance the perceptual capabilities of highly mobile legged platforms through haptic perception and active exploration. In this light, THING will deliver: 1) Novel foot designs for enhanced tactile perception and locomotion, 2) Improved perceptual capability, enriching existing modalities (lidar, vision) with haptic information, 3) Heightened physical sense of the environment, including friction, ground stability (difficult through vision alone), and 4) Enhanced mobility through improved perception, prediction, and control.

The project's first 18 months have focused on the development of: 1) novel mechatronic designs and prototyping of adaptive feet and ankles for the ANYmal quadruped, 2) new localisation and navigation algorithms fusing haptic information with traditional exteroceptive perception, 3) novel control methods for physical interaction and haptic sensing of the environment, 4) novel planning and optimisation approaches to leverage whole-body impedance and adaptive feet, and 5) the preliminary testing and evaluation of our progress in situ real-world subterranean mines and sewers.
* UNIPI designed and released the first prototypes of the THING passive adaptive foot. Inspired by the anatomy of the human foot, this unique foot present a novel way to sense contact forces, detect slippages, and conform to underlying terrains.
* UNIPI also designed and released the first prototype of the THING active adaptive foot. A similar articulated design to the passive adaptive foot, but with an actuated degree-of-freedom to enable grasping and pre-shaping.
* ETH released prototypes of a passive ankle with rigid sole. Designed to be robust to absorb shocks during locomotion while increasing contact surface area for increased traction.
* ETH also designed an active ankle system to allow for orientation control of the rigid sole.
* ETH developed a low-cost, lightweight and IP67 sealed, force-torque sensor (which may be integrated with the new ankle joint design). ETH are commercialising this technology via their spinout BOTA Systems.
* PUT and ETH developed novel machine learning methods for terrain classification from force/torque sensing in the foot.
* UEDIN has developed a controller for active probing of the ground or surrounding walls for haptic sensing of external geometry.
* ETH has developed a probabilistic slippage detector, which is crucial for stable locomotion and recovery behaviours.
* UOXF has developed VILENS: Visual Inertial LEgged Navigation System. Because of VILENS’ tight integration of visual features with inertial and kinematics, it provides significantly lower drift during state estimation.
* UOXF and UEDIN are developing methods for haptic localisation, leveraging contact points from the legs to constrain the robot’s estimated trajectory.
* ETH has also developed a method to fuse haptic and exteroceptive mapping to estimate underlying terrain topography (e.g. of dirty water or high grass).
* ETH and ANYbotics in collaboration with UNIPI and QB have distributed a simulation environment for the ANYmal robot including new ankle joints and adaptive feet.
* ETH has also developed and released a new physics engine Raisim, which outperforms existing simulator pipelines in terms of speed and accuracy.
* UEDIN has developed a novel online semi-parametric dynamics model learning method. The algorithm combines inertial parameter adaption with online data-driven regression.
* ETH has developed a new model-predictive control framework that incorporates inherent bandwidth limits into motion planning.
* ETH developed a trajectory optimiser to discover stepping motions (including sliding contacts) without a predefined contact schedule.
* ETH has also developed two-stage Bayesian optimisation framework to learn contact schedule selection.
* UEDIN has developed a fast, dynamically stable online trajectory optimisation framework. Given footstep timings, the method is able to generate dynamically stable footstep locations and base trajectories.
* UEDIN has developed a whole-body impedance controller, considering orientation of the feet. This is critical when considering foot interaction control with a rigid sole or adaptive foot.
* UEDIN has also developed a novel, provably stable variable impedance controller, based on fractal attractor dynamics.
* UOXF has developed a hierarchical approach to legged locomotion planning that can leverage multiple controllers and sub-planners.
* PUT has developed a foothold selection algorithm that uses a Convolutional Neural Network to select optimal footholds given a local elevation map. The method will be critical for foothold selection of the new adaptive feet.
* ETH has developed a novel self-supervised learning method to predict foothold quality. The method allows the robot to relate haptic feedback to previous visual information of the terrain.
* ETH has developed a fast DDP based MPC algorithm that can achieve onboard whole-body planning for ANYmal with a 1 second horizon. They have applied this planner to a ball-balancing robot with manipulator
The second half of the project will continue to focus on the integration and exploitation of our new technology. The new feet and ankles will be integrated on additional platforms throughout the consortium and will be more extensively evaluated. Perception, control and planning algorithms will continue to be developed and improved and will also explicitly exploit the capabilities of the new foot designs. In the coming months, we plan for more challenging tests in real subterranean environments, including more difficult terrain and perception. Working with our industrial partners, we will further explore the practical uses for our technology, including industrial inspection tasks. In addition to continued scientific dissemination, we will be seeking additional commercialisation opportunities, particularly for the project's new mechatronic designs.

By the end of the project, we will have advanced the robotisation of underground inspection tasks to at least TRL 6 with a system prototype demonstration in a relevant environment. In our case, the relevant environment is not a laboratory setting, but rather real copper mines and sewage tunnels. The socio-economic and wider societal implications are that both industries and municipalities may reduce maintenance costs, improve human safety by reducing time spent underground by maintenance staff, and further increase the reliability and efficacy of the inspection data. We also will strengthen Europe’s position in the global marketplace through support of SMEs and commercialisation of developed technology into new markets.
The ANYmal robot with new adaptive feet